{"cells":[{"metadata":{"id":"43PFWJBa4IyB","colab_type":"text"},"cell_type":"markdown","source":"## Readme\n- Original notebook : Keras EfficientNet B3 Training + Inference\n\n    https://www.kaggle.com/rsmits/keras-efficientnet-b3-training-inference\n\n- Customized for : Training (Model save/load), Data augmentation(Gridmask, Augmix)\n\n    https://github.com/RobinSmits/KaggleBengaliAIHandwrittenGraphemeClassification : code for training keras model\n\n    https://keras.io/getting-started/faq/ : keras FAQ for save/load model(.h5)\n\n    https://keras.io/ko/callbacks/#modelcheckpoint : callback function for saving model\n\n- Packages : tensorflow(1.15.0->2.1.0), keras(2.2.5->2.3.1), efficientnet==1.0.0, iterative-stratification==0.1.6\n    "},{"metadata":{"trusted":true},"cell_type":"code","source":"'''\nimport os\nos.chdir('/kaggle/input/train-128')\n!pwd\n'''","execution_count":null,"outputs":[]},{"metadata":{"id":"IkVQ_PF6X0gK","colab_type":"code","colab":{},"trusted":true},"cell_type":"code","source":"'''\n!unzip train_128_0.zip\n!unzip train_128_1.zip\n!unzip train_128_2.zip\n!unzip train_128_3.zip\n'''","execution_count":null,"outputs":[]},{"metadata":{"id":"9KY4XdhQRQzV","colab_type":"code","outputId":"1ef0e857-72fa-421e-8078-109e7d216217","colab":{"base_uri":"https://localhost:8080/","height":34},"trusted":true},"cell_type":"code","source":"import os\n\nos.chdir('/kaggle/input/iterative-stratification')\n!pip install './iterative_stratification-0.1.6-py3-none-any.whl'\n\nos.chdir('/kaggle/input/kagglebengaliaihandwrittengraphemeclassification/KaggleKernelEfficientNetB3')\n!pip install './efficientnet-1.0.0-py3-none-any.whl'\n\n!pwd","execution_count":null,"outputs":[]},{"metadata":{"id":"tjGLC1skJvsA","colab_type":"text"},"cell_type":"markdown","source":"This competition provides a lot of room for interresting experimentations. In this kernel I use a rather easy way to train a standard EfficientNet B3 model with a custom head layer and Generalized mean pool. I use only basic image preprocessing with a scaling factor.\n\nTo save on training time I use a different training set on each epoch. This gives a nice boost of about 0.005 to 0.008 compared to a fixed training set when using train/test split or cross-validation. The downside is that the validation has some less value.\n\nThis kernel contains the inference part where I use 3 models from the training. For the complete code to train it yourself you can download it from my [github](https://github.com/RobinSmits/KaggleBengaliAIHandwrittenGraphemeClassification). I trained it for 80 epochs on my 1070 Ti (roughly 1,5 days).\n\nI hope you like it and if you find this kernel helpfull..then please don't forget to upvote it."},{"metadata":{"id":"2fbFDufwJvsD","colab_type":"code","outputId":"d959ea5d-a0f7-4154-9ddb-a7c4d7e18e4e","colab":{"base_uri":"https://localhost:8080/","height":34},"trusted":true},"cell_type":"code","source":"import cv2\nimport os\nimport time, gc\nimport numpy as np\nimport pandas as pd\n\nimport tensorflow as tf\nimport keras\nfrom keras import backend as K\nfrom keras.models import Model, Input\nfrom keras.layers import Dense, Lambda\nfrom math import ceil\n\nimport efficientnet.keras as efn","execution_count":null,"outputs":[]},{"metadata":{"id":"wv9oHNL-WWOC","colab_type":"code","colab":{},"trusted":true},"cell_type":"code","source":"import math\nimport random\nimport warnings\nfrom PIL import Image\nfrom glob import glob\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nimport tqdm\nimport efficientnet.keras as efn\n#sklearns \nfrom sklearn.metrics import cohen_kappa_score, accuracy_score\nfrom sklearn.model_selection import train_test_split \n\n# keras modules \nfrom keras.applications.densenet import DenseNet121, DenseNet169, DenseNet201\nfrom keras.optimizers import Adam, Nadam, SGD\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom keras.models import Model, load_model, Sequential\nfrom keras.layers import Dense, GlobalAveragePooling2D, Dropout, Conv2D, GlobalMaxPooling2D, concatenate\nfrom keras.layers import (MaxPooling2D, Input, Average, Activation, MaxPool2D,\n                          Flatten, LeakyReLU, BatchNormalization)\nfrom keras import models\nfrom keras import layers\nfrom keras.preprocessing.image import load_img\nfrom keras.preprocessing.image import img_to_array\n\nfrom keras.utils import Sequence\nfrom keras import utils as np_utils\nfrom keras.callbacks import (Callback, ModelCheckpoint,\n                                        LearningRateScheduler,EarlyStopping, \n                                        ReduceLROnPlateau,CSVLogger)\n\n# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load in \n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\nfrom tqdm.auto import tqdm\nfrom glob import glob\nfrom keras.utils import Sequence\n\nfrom tensorflow import keras\nimport matplotlib.image as mpimg\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom keras.models import Model\nfrom keras.models import clone_model\nfrom keras.layers import Dense,Conv2D,Flatten,MaxPool2D,Dropout,BatchNormalization, Input\nfrom keras.optimizers import Adam\nfrom keras.callbacks import ReduceLROnPlateau\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import confusion_matrix\nimport PIL.Image as Image, PIL.ImageDraw as ImageDraw, PIL.ImageFont as ImageFont\nfrom matplotlib import pyplot as plt\nimport seaborn as sns\n\n# Input data files are available in the \"../input/\" directory.\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\n#import os\n#for dirname, _, filenames in os.walk('/kaggle/input'):\n#    for filename in filenames:\n#        print(os.path.join(dirname, filename))\n\n# Import Modules\nimport time, gc\nfrom math import floor\n# Keras\nimport keras.backend as K\nfrom keras.optimizers import Adam\nfrom keras.callbacks import Callback, ModelCheckpoint\n\n# Iterative-Stratification\nfrom iterstrat.ml_stratifiers import MultilabelStratifiedKFold, MultilabelStratifiedShuffleSplit\nos.chdir('/kaggle/input/kagglebengaliaihandwrittengraphemeclassification/KaggleKernelEfficientNetB3')\n# Custom \nfrom preprocessing import generate_images, resize_image\nfrom model import create_model\nfrom utils import plot_summaries","execution_count":null,"outputs":[]},{"metadata":{"id":"iDDoWKhcJvsI","colab_type":"code","colab":{},"trusted":true},"cell_type":"code","source":"# Constants\nHEIGHT = 137\nWIDTH = 236\nFACTOR = 0.70\nHEIGHT_NEW = 128 #int(HEIGHT * FACTOR)\nWIDTH_NEW = 128 #int(WIDTH * FACTOR)\nCHANNELS = 3\nBATCH_SIZE = 16\n\n# Dir\nDIR = '/kaggle/input/bengaliai-cv19' #'../input/bengaliai-cv19'","execution_count":null,"outputs":[]},{"metadata":{"id":"JQCJCNPdXMVr","colab_type":"code","colab":{},"trusted":true},"cell_type":"code","source":"train_df_ = pd.read_csv('/kaggle/input/bengaliai-cv19/train.csv')\ntest_df_ = pd.read_csv('/kaggle/input/bengaliai-cv19/test.csv')\nclass_map_df = pd.read_csv('/kaggle/input/bengaliai-cv19/class_map.csv')\nsample_sub_df = pd.read_csv('/kaggle/input/bengaliai-cv19/sample_submission.csv')","execution_count":null,"outputs":[]},{"metadata":{"id":"-71G-6I8Ttt7","colab_type":"text"},"cell_type":"markdown","source":"## Data Augmentation (Gridmask, Augmix)"},{"metadata":{"id":"HDJpIj7RTezW","colab_type":"code","colab":{},"trusted":true},"cell_type":"code","source":"SEED = 2020\nbatch_size = 12 \n# dim, size는 64x64 라서 64로 수정\ndim = (128, 128)\nSIZE = 128\nstats = (0.0692, 0.2051)\nHEIGHT = 137 \nWIDTH = 236\n\nimport random\ndef seed_all(SEED):\n    random.seed(SEED)\n    np.random.seed(SEED)\n    os.environ['PYTHONHASHSEED'] = str(SEED)\n    \n# seed all\nseed_all(SEED)","execution_count":null,"outputs":[]},{"metadata":{"id":"IymmG2PxT4Qk","colab_type":"code","colab":{},"trusted":true},"cell_type":"code","source":"import os\nimport cv2\nimport numpy as np\nimport pandas as pd\nimport albumentations\nfrom albumentations.core.transforms_interface import DualTransform\nfrom albumentations.augmentations import functional as F\nimport matplotlib.pyplot as plt\nimport torch\nfrom torch.utils.data import TensorDataset, DataLoader, Dataset\n\n## Grid Mask\n# code takesn from https://www.kaggle.com/haqishen/gridmask\n\nimport albumentations\nfrom albumentations.core.transforms_interface import DualTransform, ImageOnlyTransform\nfrom albumentations.augmentations import functional as F\n\nclass GridMask(DualTransform):\n    \"\"\"GridMask augmentation for image classification and object detection.\n\n    Args:\n        num_grid (int): number of grid in a row or column.\n        fill_value (int, float, lisf of int, list of float): value for dropped pixels.\n        rotate ((int, int) or int): range from which a random angle is picked. If rotate is a single int\n            an angle is picked from (-rotate, rotate). Default: (-90, 90)\n        mode (int):\n            0 - cropout a quarter of the square of each grid (left top)\n            1 - reserve a quarter of the square of each grid (left top)\n            2 - cropout 2 quarter of the square of each grid (left top & right bottom)\n\n    Targets:\n        image, mask\n\n    Image types:\n        uint8, float32\n\n    Reference:\n    |  https://arxiv.org/abs/2001.04086\n    |  https://github.com/akuxcw/GridMask\n    \"\"\"\n\n    def __init__(self, num_grid=3, fill_value=0, rotate=0, mode=0, always_apply=False, p=0.5):\n        super(GridMask, self).__init__(always_apply, p)\n        if isinstance(num_grid, int):\n            num_grid = (num_grid, num_grid)\n        if isinstance(rotate, int):\n            rotate = (-rotate, rotate)\n        self.num_grid = num_grid\n        self.fill_value = fill_value\n        self.rotate = rotate\n        self.mode = mode\n        self.masks = None\n        self.rand_h_max = []\n        self.rand_w_max = []\n\n    def init_masks(self, height, width):\n        if self.masks is None:\n            self.masks = []\n            n_masks = self.num_grid[1] - self.num_grid[0] + 1\n            for n, n_g in enumerate(range(self.num_grid[0], self.num_grid[1] + 1, 1)):\n                grid_h = height / n_g\n                grid_w = width / n_g\n                this_mask = np.ones((int((n_g + 1) * grid_h), int((n_g + 1) * grid_w))).astype(np.uint8)\n                for i in range(n_g + 1):\n                    for j in range(n_g + 1):\n                        this_mask[\n                             int(i * grid_h) : int(i * grid_h + grid_h / 2),\n                             int(j * grid_w) : int(j * grid_w + grid_w / 2)\n                        ] = self.fill_value\n                        if self.mode == 2:\n                            this_mask[\n                                 int(i * grid_h + grid_h / 2) : int(i * grid_h + grid_h),\n                                 int(j * grid_w + grid_w / 2) : int(j * grid_w + grid_w)\n                            ] = self.fill_value\n                \n                if self.mode == 1:\n                    this_mask = 1 - this_mask\n\n                self.masks.append(this_mask)\n                self.rand_h_max.append(grid_h)\n                self.rand_w_max.append(grid_w)\n\n    def apply(self, image, mask, rand_h, rand_w, angle, **params):\n        h, w = image.shape[:2]\n        mask = F.rotate(mask, angle) if self.rotate[1] > 0 else mask\n        mask = mask[:,:,np.newaxis] if image.ndim == 3 else mask\n        image *= mask[rand_h:rand_h+h, rand_w:rand_w+w].astype(image.dtype)\n        return image\n\n    def get_params_dependent_on_targets(self, params):\n        img = params['image']\n        height, width = img.shape[:2]\n        self.init_masks(height, width)\n\n        mid = np.random.randint(len(self.masks))\n        mask = self.masks[mid]\n        rand_h = np.random.randint(self.rand_h_max[mid])\n        rand_w = np.random.randint(self.rand_w_max[mid])\n        angle = np.random.randint(self.rotate[0], self.rotate[1]) if self.rotate[1] > 0 else 0\n\n        return {'mask': mask, 'rand_h': rand_h, 'rand_w': rand_w, 'angle': angle}\n\n    @property\n    def targets_as_params(self):\n        return ['image']\n\n    def get_transform_init_args_names(self):\n        return ('num_grid', 'fill_value', 'rotate', 'mode')","execution_count":null,"outputs":[]},{"metadata":{"id":"LDGZeEI8T53x","colab_type":"code","colab":{},"trusted":true},"cell_type":"code","source":"# augmix : https://github.com/google-research/augmix\n\nfrom PIL import Image\nfrom PIL import ImageOps\nimport numpy as np\n\ndef int_parameter(level, maxval):\n    \"\"\"Helper function to scale `val` between 0 and maxval .\n    Args:\n    level: Level of the operation that will be between [0, `PARAMETER_MAX`].\n    maxval: Maximum value that the operation can have. This will be scaled to\n      level/PARAMETER_MAX.\n    Returns:\n    An int that results from scaling `maxval` according to `level`.\n    \"\"\"\n    return int(level * maxval / 10)\n\n\ndef float_parameter(level, maxval):\n    \"\"\"Helper function to scale `val` between 0 and maxval.\n    Args:\n    level: Level of the operation that will be between [0, `PARAMETER_MAX`].\n    maxval: Maximum value that the operation can have. This will be scaled to\n      level/PARAMETER_MAX.\n    Returns:\n    A float that results from scaling `maxval` according to `level`.\n    \"\"\"\n    return float(level) * maxval / 10.\n\ndef sample_level(n):\n    return np.random.uniform(low=0.1, high=n)\n\ndef autocontrast(pil_img, _):\n    return ImageOps.autocontrast(pil_img)\n\ndef equalize(pil_img, _):\n    return ImageOps.equalize(pil_img)\n\ndef posterize(pil_img, level):\n    level = int_parameter(sample_level(level), 4)\n    return ImageOps.posterize(pil_img, 4 - level)\n\ndef rotate(pil_img, level):\n    degrees = int_parameter(sample_level(level), 30)\n    if np.random.uniform() > 0.5:\n        degrees = -degrees\n    return pil_img.rotate(degrees, resample=Image.BILINEAR)\n\ndef solarize(pil_img, level):\n    level = int_parameter(sample_level(level), 256)\n    return ImageOps.solarize(pil_img, 256 - level)\n\ndef shear_x(pil_img, level):\n    level = float_parameter(sample_level(level), 0.3)\n    if np.random.uniform() > 0.5:\n        level = -level\n    return pil_img.transform((SIZE, SIZE),\n                           Image.AFFINE, (1, level, 0, 0, 1, 0),\n                           resample=Image.BILINEAR)\n\ndef shear_y(pil_img, level):\n    level = float_parameter(sample_level(level), 0.3)\n    if np.random.uniform() > 0.5:\n        level = -level\n    return pil_img.transform((SIZE, SIZE),\n                           Image.AFFINE, (1, 0, 0, level, 1, 0),\n                           resample=Image.BILINEAR)\n\ndef translate_x(pil_img, level):\n    level = int_parameter(sample_level(level), SIZE / 3)\n    if np.random.random() > 0.5:\n        level = -level\n    return pil_img.transform((SIZE, SIZE),\n                           Image.AFFINE, (1, 0, level, 0, 1, 0),\n                           resample=Image.BILINEAR)\n\n\ndef translate_y(pil_img, level):\n    level = int_parameter(sample_level(level), SIZE / 3)\n    if np.random.random() > 0.5:\n        level = -level\n    return pil_img.transform((SIZE, SIZE),\n                           Image.AFFINE, (1, 0, 0, 0, 1, level),\n                           resample=Image.BILINEAR)\n\naugmentations = [\n    autocontrast, equalize, posterize, rotate, solarize, shear_x, shear_y,\n    translate_x, translate_y\n]\n\n# taken from https://www.kaggle.com/iafoss/image-preprocessing-128x128\nMEAN = [ 0.06922848809290576,  0.06922848809290576,  0.06922848809290576]\nSTD = [ 0.20515700083327537,  0.20515700083327537,  0.20515700083327537]\n\ndef normalize(image):\n    \"\"\"Normalize input image channel-wise to zero mean and unit variance.\"\"\"\n    image = image.transpose(2, 0, 1)  # Switch to channel-first\n    mean, std = np.array(MEAN), np.array(STD)\n    image = (image - mean[:, None, None]) / std[:, None, None]\n    return image.transpose(1, 2, 0)\n\n\ndef apply_op(image, op, severity):\n    image = np.clip(image * 255., 0, 255).astype(np.uint8)\n    pil_img = Image.fromarray(image)  # Convert to PIL.Image\n    pil_img = op(pil_img, severity)\n    return np.asarray(pil_img) / 255.\n\n\ndef augment_and_mix(image, severity=1, width=3, depth=1, alpha=1.):\n    \"\"\"Perform AugMix augmentations and compute mixture.\n    Args:\n    image: Raw input image as float32 np.ndarray of shape (h, w, c)\n    severity: Severity of underlying augmentation operators (between 1 to 10).\n    width: Width of augmentation chain\n    depth: Depth of augmentation chain. -1 enables stochastic depth uniformly\n      from [1, 3]\n    alpha: Probability coefficient for Beta and Dirichlet distributions.\n    Returns:\n    mixed: Augmented and mixed image.\n  \"\"\"\n    ws = np.float32(\n      np.random.dirichlet([alpha] * width))\n    m = np.float32(np.random.beta(alpha, alpha))\n\n    mix = np.zeros_like(image)\n    for i in range(width):\n        image_aug = image.copy()\n        depth = depth if depth > 0 else np.random.randint(1, 4)\n        \n        for _ in range(depth):\n            op = np.random.choice(augmentations)\n            image_aug = apply_op(image_aug, op, severity)\n        mix = np.add(mix, ws[i] * normalize(image_aug), out=mix, \n                     casting=\"unsafe\")\n\n    mixed = (1 - m) * normalize(image) + m * mix\n    return mixed","execution_count":null,"outputs":[]},{"metadata":{"id":"uWZj94aFJvsM","colab_type":"text"},"cell_type":"markdown","source":"## Image Preprocessing"},{"metadata":{"id":"NLteAVN_JvsN","colab_type":"code","outputId":"93663f4a-3050-4950-9efd-a6bf85087da6","colab":{"base_uri":"https://localhost:8080/","height":51},"trusted":true},"cell_type":"code","source":"'''\n# Image Size Summary\nprint(HEIGHT_NEW)\nprint(WIDTH_NEW)\n\n# Image Prep\ndef resize_image(img, WIDTH_NEW, HEIGHT_NEW):\n    # Invert\n    img = 255 - img\n\n    # Normalize\n    img = (img * (255.0 / img.max())).astype(np.uint8)\n\n    # Reshape\n    img = img.reshape(HEIGHT, WIDTH)\n    image_resized = cv2.resize(img, (WIDTH_NEW, HEIGHT_NEW), interpolation = cv2.INTER_AREA)\n\n    return image_resized.reshape(-1)\n'''   ","execution_count":null,"outputs":[]},{"metadata":{"id":"ZYHmbEZsJvsQ","colab_type":"text"},"cell_type":"markdown","source":"## Create Model"},{"metadata":{"id":"akccWV2zJvsR","colab_type":"code","colab":{},"trusted":true},"cell_type":"code","source":"# Generalized mean pool - GeM\ngm_exp = tf.Variable(3.0, dtype = tf.float32)\ndef generalized_mean_pool_2d(X):\n    pool = (tf.reduce_mean(tf.abs(X**(gm_exp)),\n                        axis = [1, 2], \n                        keepdims = False) + 1.e-7)**(1./gm_exp)\n    return pool","execution_count":null,"outputs":[]},{"metadata":{"colab_type":"code","id":"gVKlF1x4WC4s","colab":{},"trusted":true},"cell_type":"code","source":"from keras.models import model_from_json\n\n# Create Model\ndef create_model(input_shape):\n    '''\n    # Input Layer\n    input = Input(shape = input_shape)\n    \n    # Create and Compile Model and show Summary\n    x_model = efn.EfficientNetB3(weights = None, include_top = False, input_tensor = input, pooling = None, classes = None)\n    \n    # UnFreeze all layers\n    for layer in x_model.layers:\n        layer.trainable = True\n    \n    # GeM\n    lambda_layer = Lambda(generalized_mean_pool_2d)\n    lambda_layer.trainable_weights.extend([gm_exp])\n    x = lambda_layer(x_model.output)\n    \n    # multi output\n    grapheme_root = Dense(168, activation = 'softmax', name = 'root')(x)\n    vowel_diacritic = Dense(11, activation = 'softmax', name = 'vowel')(x)\n    consonant_diacritic = Dense(7, activation = 'softmax', name = 'consonant')(x)\n\n    # model\n    model = Model(inputs = x_model.input, outputs = [grapheme_root, vowel_diacritic, consonant_diacritic])\n    '''\n    os.chdir('/kaggle/input/json-string')\n    model = model_from_json(json_string)\n    return model","execution_count":null,"outputs":[]},{"metadata":{"id":"X65l9criWhFD","colab_type":"text"},"cell_type":"markdown","source":"## Training (train.py)"},{"metadata":{"id":"WKJ30QddWh5u","colab_type":"code","outputId":"9f72d751-bc20-437a-d5a9-85c8150f5096","colab":{"base_uri":"https://localhost:8080/","height":1000},"trusted":true},"cell_type":"code","source":"os.chdir('/kaggle/working')\n\n# Import Modules\nimport os\nimport time, gc\nimport numpy as np\nimport pandas as pd\nfrom math import floor\nimport cv2\nimport tensorflow as tf\n\n# Keras\nimport keras\nimport keras.backend as K\nfrom keras.optimizers import Adam\nfrom keras.callbacks import Callback, ModelCheckpoint\n\n# Iterative-Stratification\nfrom iterstrat.ml_stratifiers import MultilabelStratifiedKFold, MultilabelStratifiedShuffleSplit\n\n# Custom \nfrom preprocessing import generate_images, resize_image\nfrom model import create_model\nfrom utils import plot_summaries\n\n# Seeds\nSEED = 1234\nnp.random.seed(SEED)\ntf.random.set_seed(SEED)\n\n# Input Dir\nDATA_DIR = '/kaggle/input/bengaliai-cv19'     #'C:/KaggleBengaliAI/bengaliai-cv19'\nTRAIN_DIR = '/kaggle/input/train-128-128/train_128_128/'         #'./train/'\n\n# Constants\nHEIGHT = 137\nWIDTH = 236\nSCALE_FACTOR = 0.70\nHEIGHT_NEW = 128 #int(HEIGHT * SCALE_FACTOR)\nWIDTH_NEW = 128 #int(WIDTH * SCALE_FACTOR)\nRUN_NAME = 'Train2_'\nPLOT_NAME1 = 'Train1_LossAndAccuracy.png'\nPLOT_NAME2 = 'Train1_Recall.png'\n\nBATCH_SIZE = 56\nCHANNELS = 3\nEPOCHS = 2\nTEST_SIZE = 1./6\n\n# Image Size Summary\n#print(HEIGHT_NEW)\n#print(WIDTH_NEW)\n\n# Generate Image (Has to be done only one time .. or again when changing SCALE_FACTOR)\n#GENERATE_IMAGES = True\n#if GENERATE_IMAGES:\n#    generate_images(DATA_DIR, TRAIN_DIR, WIDTH, HEIGHT, WIDTH_NEW, HEIGHT_NEW)\n\n# Prepare Train Labels (Y)\ntrain_df = pd.read_csv(os.path.join(DATA_DIR, 'train.csv'))\ntgt_cols = ['grapheme_root', 'vowel_diacritic', 'consonant_diacritic']\ndesc_df = train_df[tgt_cols].astype('str').describe()\ntypes = desc_df.loc['unique',:]\nX_train = train_df['image_id'].values\ntrain_df = train_df[tgt_cols].astype('uint8')\nfor col in tgt_cols:\n    train_df[col] = train_df[col].map('{:03}'.format)\nY_train = pd.get_dummies(train_df)\n\n# Cleanup\ndel train_df\ngc.collect()\n\n# Modelcheckpoint\ndef ModelCheckpointFull(model_name):\n    return ModelCheckpoint(model_name, \n                            monitor = 'val_loss', \n                            verbose = 1, \n                            save_best_only = False, \n                            save_weights_only = True, \n                            mode = 'min', \n                            period = 1)\n\n\ndef _read(path):\n    img = cv2.imread(path)    \n    return img\n\nclass TrainDataGenerator(keras.utils.Sequence):\n#    def __init__(self, X_set, Y_set, ids, batch_size = 16, img_size = (512, 512, 3), img_dir = TRAIN_DIR, *args, **kwargs):\n    def __init__(self, X_set, Y_set, ids, batch_size = 16, img_size = (128, 128, 3), img_dir = TRAIN_DIR, transform=None):\n        self.X = X_set\n        self.ids = ids\n        self.Y = Y_set\n        self.batch_size = batch_size\n        self.img_size = img_size\n        self.img_dir = img_dir\n        self.on_epoch_end()\n        self.transform = transform\n\n        # Split Data\n        self.x_indexed = self.X[self.ids]\n        self.y_indexed = self.Y.iloc[self.ids]\n\n        # Prep Y per Label   \n        self.y_root = self.y_indexed.iloc[:,0:types['grapheme_root']].values\n        self.y_vowel = self.y_indexed.iloc[:,types['grapheme_root']:types['grapheme_root']+types['vowel_diacritic']].values\n        self.y_consonant = self.y_indexed.iloc[:,types['grapheme_root']+types['vowel_diacritic']:].values\n    \n    def __len__(self):\n        return int(floor(len(self.ids) / self.batch_size))\n\n    def __getitem__(self, index):\n        indices = self.indices[index*self.batch_size:(index+1)*self.batch_size]\n        X, Y_root, Y_vowel, Y_consonant = self.__data_generation(indices)\n        return X, {'root': Y_root, 'vowel': Y_vowel, 'consonant': Y_consonant}\n\n    def on_epoch_end(self):\n        self.indices = np.arange(len(self.ids))\n    \n    def __data_generation(self, indices):\n        X = np.empty((self.batch_size, *self.img_size))\n        Y_root = np.empty((self.batch_size, 168), dtype = np.int16)\n        Y_vowel = np.empty((self.batch_size, 11), dtype = np.int16)\n        Y_consonant = np.empty((self.batch_size, 7), dtype = np.int16)\n\n        # Get Images for Batch\n        for i, index in enumerate(indices):\n            ID = self.x_indexed[index]\n            image = _read(self.img_dir+ID+\".png\")\n            image = cv2.resize(image,(128,128))#self._dim) \n            \n            if self.transform is not None:\n                if np.random.rand() > 0.7:\n                    # albumentation : grid mask\n                    res = self.transform(image=image)\n                    image = res['image']\n                else:\n                    # augmix augmentation\n                    image = augment_and_mix(image)\n            \n            # scaling  이거 지우면 scaling 안한 파일이 들어감\n            image = (image.astype(np.float32)/255.0 - stats[0])/stats[1]\n            \n            # gray scaling 128x128\n            gray = lambda rgb : np.dot(rgb[... , :3] , [0.299 , 0.587, 0.114]) \n            image = gray(image) \n            \n            # expand the axises , 128x128x3 만들기\n            \n            image = image.reshape(128,128,1)\n            image = np.concatenate((image,)*3, axis=-1)\n\n\n            X[i,] = image\n        \n        # Get Labels for Batch\n        Y_root = self.y_root[indices]\n        Y_vowel = self.y_vowel[indices]\n        Y_consonant = self.y_consonant[indices]    \n       \n        return X, Y_root, Y_vowel, Y_consonant \n\n# Create Model\n#model = create_model(input_shape = (HEIGHT_NEW, WIDTH_NEW, CHANNELS))\nimport json\nfrom keras.models import model_from_json\nmodel_json = \"{\\\"class_name\\\": \\\"Model\\\", \\\"config\\\": {\\\"name\\\": \\\"efficientnet-b3\\\", \\\"layers\\\": [{\\\"name\\\": \\\"input_5\\\", \\\"class_name\\\": \\\"InputLayer\\\", \\\"config\\\": {\\\"batch_input_shape\\\": [null, 128, 128, 3], \\\"dtype\\\": \\\"float32\\\", \\\"sparse\\\": false, \\\"name\\\": \\\"input_5\\\"}, \\\"inbound_nodes\\\": []}, {\\\"name\\\": \\\"stem_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"stem_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 40, \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [2, 2], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"input_5\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"stem_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"stem_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"stem_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"stem_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"stem_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"stem_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"stem_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1a_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block1a_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block1a_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 40]}, \\\"inbound_nodes\\\": [[[\\\"block1a_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 10, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1a_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 40, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1a_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block1a_activation\\\", 0, 0, {}], [\\\"block1a_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 24, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1a_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1a_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block1a_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1a_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1b_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block1b_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block1b_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 24]}, \\\"inbound_nodes\\\": [[[\\\"block1b_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 6, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1b_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 24, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1b_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block1b_activation\\\", 0, 0, {}], [\\\"block1b_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 24, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1b_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1b_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.0125, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1b_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block1b_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block1b_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block1b_drop\\\", 0, 0, {}], [\\\"block1a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 144, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block1b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2a_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2a_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [2, 2], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2a_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2a_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2a_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2a_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 144]}, \\\"inbound_nodes\\\": [[[\\\"block2a_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 6, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2a_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 144, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2a_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2a_activation\\\", 0, 0, {}], [\\\"block2a_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 32, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2a_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2a_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block2a_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2a_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 192, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2b_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2b_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2b_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2b_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2b_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2b_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 192]}, \\\"inbound_nodes\\\": [[[\\\"block2b_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 8, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2b_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 192, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2b_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2b_activation\\\", 0, 0, {}], [\\\"block2b_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 32, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2b_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2b_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.037500000000000006, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2b_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2b_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block2b_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2b_drop\\\", 0, 0, {}], [\\\"block2a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 192, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2c_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2c_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2c_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2c_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2c_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2c_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 192]}, \\\"inbound_nodes\\\": [[[\\\"block2c_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 8, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2c_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 192, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2c_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2c_activation\\\", 0, 0, {}], [\\\"block2c_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 32, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2c_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2c_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.05, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2c_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block2c_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block2c_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block2c_drop\\\", 0, 0, {}], [\\\"block2b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 192, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block2c_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3a_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3a_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [2, 2], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3a_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3a_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3a_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3a_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 192]}, \\\"inbound_nodes\\\": [[[\\\"block3a_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 8, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3a_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 192, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3a_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3a_activation\\\", 0, 0, {}], [\\\"block3a_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 48, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3a_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3a_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block3a_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3a_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 288, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3b_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3b_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3b_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3b_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3b_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3b_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 288]}, \\\"inbound_nodes\\\": [[[\\\"block3b_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 12, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3b_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 288, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3b_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3b_activation\\\", 0, 0, {}], [\\\"block3b_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 48, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3b_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3b_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.07500000000000001, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3b_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3b_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block3b_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3b_drop\\\", 0, 0, {}], [\\\"block3a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 288, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3c_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3c_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3c_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3c_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3c_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3c_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 288]}, \\\"inbound_nodes\\\": [[[\\\"block3c_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 12, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3c_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 288, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3c_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3c_activation\\\", 0, 0, {}], [\\\"block3c_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 48, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3c_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3c_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.08750000000000001, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3c_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block3c_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block3c_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block3c_drop\\\", 0, 0, {}], [\\\"block3b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 288, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block3c_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4a_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4a_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [2, 2], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4a_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4a_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4a_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4a_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 288]}, \\\"inbound_nodes\\\": [[[\\\"block4a_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 12, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4a_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 288, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4a_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4a_activation\\\", 0, 0, {}], [\\\"block4a_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 96, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4a_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4a_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4a_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4a_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4b_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4b_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4b_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4b_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4b_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4b_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 576]}, \\\"inbound_nodes\\\": [[[\\\"block4b_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 24, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4b_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4b_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4b_activation\\\", 0, 0, {}], [\\\"block4b_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 96, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4b_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4b_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.1125, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4b_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4b_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block4b_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4b_drop\\\", 0, 0, {}], [\\\"block4a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4c_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4c_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4c_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4c_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4c_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4c_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 576]}, \\\"inbound_nodes\\\": [[[\\\"block4c_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 24, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4c_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4c_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4c_activation\\\", 0, 0, {}], [\\\"block4c_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 96, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4c_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4c_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.125, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4c_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4c_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block4c_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4c_drop\\\", 0, 0, {}], [\\\"block4b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4c_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4d_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4d_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4d_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4d_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4d_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4d_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 576]}, \\\"inbound_nodes\\\": [[[\\\"block4d_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 24, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4d_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4d_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4d_activation\\\", 0, 0, {}], [\\\"block4d_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 96, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4d_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4d_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.1375, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4d_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4d_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block4d_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4d_drop\\\", 0, 0, {}], [\\\"block4c_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4d_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4e_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4e_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4e_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4e_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4e_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4e_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 576]}, \\\"inbound_nodes\\\": [[[\\\"block4e_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 24, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4e_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4e_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4e_activation\\\", 0, 0, {}], [\\\"block4e_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 96, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4e_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4e_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.15000000000000002, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4e_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block4e_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block4e_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block4e_drop\\\", 0, 0, {}], [\\\"block4d_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block4e_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5a_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5a_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5a_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5a_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5a_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5a_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 576]}, \\\"inbound_nodes\\\": [[[\\\"block5a_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 24, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5a_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 576, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5a_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5a_activation\\\", 0, 0, {}], [\\\"block5a_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 136, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5a_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5a_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5a_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5a_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5b_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5b_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5b_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5b_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5b_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5b_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 816]}, \\\"inbound_nodes\\\": [[[\\\"block5b_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 34, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5b_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5b_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5b_activation\\\", 0, 0, {}], [\\\"block5b_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 136, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5b_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5b_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.17500000000000002, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5b_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5b_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block5b_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5b_drop\\\", 0, 0, {}], [\\\"block5a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5c_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5c_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5c_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5c_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5c_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5c_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 816]}, \\\"inbound_nodes\\\": [[[\\\"block5c_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 34, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5c_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5c_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5c_activation\\\", 0, 0, {}], [\\\"block5c_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 136, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5c_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5c_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.1875, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5c_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5c_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block5c_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5c_drop\\\", 0, 0, {}], [\\\"block5b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5c_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5d_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5d_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5d_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5d_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5d_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5d_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 816]}, \\\"inbound_nodes\\\": [[[\\\"block5d_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 34, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5d_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5d_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5d_activation\\\", 0, 0, {}], [\\\"block5d_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 136, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5d_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5d_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.2, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5d_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5d_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block5d_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5d_drop\\\", 0, 0, {}], [\\\"block5c_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5d_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5e_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5e_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5e_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5e_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5e_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5e_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 816]}, \\\"inbound_nodes\\\": [[[\\\"block5e_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 34, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5e_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5e_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5e_activation\\\", 0, 0, {}], [\\\"block5e_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 136, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5e_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5e_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.21250000000000002, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5e_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block5e_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block5e_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block5e_drop\\\", 0, 0, {}], [\\\"block5d_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block5e_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6a_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6a_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [2, 2], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6a_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6a_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6a_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6a_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 816]}, \\\"inbound_nodes\\\": [[[\\\"block6a_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 34, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6a_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 816, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6a_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6a_activation\\\", 0, 0, {}], [\\\"block6a_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 232, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6a_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6a_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6a_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6a_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6b_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6b_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6b_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6b_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6b_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6b_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 1392]}, \\\"inbound_nodes\\\": [[[\\\"block6b_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 58, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6b_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6b_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6b_activation\\\", 0, 0, {}], [\\\"block6b_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 232, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6b_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6b_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.23750000000000002, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6b_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6b_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block6b_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6b_drop\\\", 0, 0, {}], [\\\"block6a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6c_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6c_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6c_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6c_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6c_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6c_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 1392]}, \\\"inbound_nodes\\\": [[[\\\"block6c_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 58, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6c_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6c_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6c_activation\\\", 0, 0, {}], [\\\"block6c_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 232, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6c_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6c_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.25, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6c_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6c_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block6c_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6c_drop\\\", 0, 0, {}], [\\\"block6b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6c_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6d_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6d_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6d_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6d_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6d_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6d_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 1392]}, \\\"inbound_nodes\\\": [[[\\\"block6d_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 58, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6d_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6d_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6d_activation\\\", 0, 0, {}], [\\\"block6d_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 232, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6d_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6d_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.2625, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6d_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6d_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block6d_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6d_drop\\\", 0, 0, {}], [\\\"block6c_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6d_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6e_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6e_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6e_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6e_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6e_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6e_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 1392]}, \\\"inbound_nodes\\\": [[[\\\"block6e_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 58, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6e_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6e_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6e_activation\\\", 0, 0, {}], [\\\"block6e_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 232, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6e_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6e_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.275, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6e_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6e_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block6e_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6e_drop\\\", 0, 0, {}], [\\\"block6d_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6e_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6f_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6f_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [5, 5], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6f_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6f_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6f_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6f_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 1392]}, \\\"inbound_nodes\\\": [[[\\\"block6f_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 58, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6f_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6f_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6f_activation\\\", 0, 0, {}], [\\\"block6f_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 232, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6f_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6f_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.28750000000000003, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6f_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block6f_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block6f_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block6f_drop\\\", 0, 0, {}], [\\\"block6e_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block6f_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7a_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block7a_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7a_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7a_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block7a_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block7a_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 1392]}, \\\"inbound_nodes\\\": [[[\\\"block7a_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 58, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7a_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1392, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7a_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block7a_activation\\\", 0, 0, {}], [\\\"block7a_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 384, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7a_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7a_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block7a_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7a_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_expand_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_expand_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 2304, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_expand_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_expand_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7b_expand_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_expand_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_expand_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block7b_expand_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_dwconv\\\", \\\"class_name\\\": \\\"DepthwiseConv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_dwconv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"kernel_size\\\": [3, 3], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"bias_constraint\\\": null, \\\"depth_multiplier\\\": 1, \\\"depthwise_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"depthwise_regularizer\\\": null, \\\"depthwise_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7b_expand_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7b_dwconv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"block7b_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_se_squeeze\\\", \\\"class_name\\\": \\\"GlobalAveragePooling2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_se_squeeze\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"data_format\\\": \\\"channels_last\\\"}, \\\"inbound_nodes\\\": [[[\\\"block7b_activation\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_se_reshape\\\", \\\"class_name\\\": \\\"Reshape\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_se_reshape\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"target_shape\\\": [1, 1, 2304]}, \\\"inbound_nodes\\\": [[[\\\"block7b_se_squeeze\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_se_reduce\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_se_reduce\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 96, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"swish\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7b_se_reshape\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_se_expand\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_se_expand\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 2304, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"sigmoid\\\", \\\"use_bias\\\": true, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7b_se_reduce\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_se_excite\\\", \\\"class_name\\\": \\\"Multiply\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_se_excite\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block7b_activation\\\", 0, 0, {}], [\\\"block7b_se_expand\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_project_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_project_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 384, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7b_se_excite\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_project_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_project_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7b_project_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_drop\\\", \\\"class_name\\\": \\\"FixedDropout\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_drop\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"rate\\\": 0.3125, \\\"noise_shape\\\": [null, 1, 1, 1], \\\"seed\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7b_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"block7b_add\\\", \\\"class_name\\\": \\\"Add\\\", \\\"config\\\": {\\\"name\\\": \\\"block7b_add\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\"}, \\\"inbound_nodes\\\": [[[\\\"block7b_drop\\\", 0, 0, {}], [\\\"block7a_project_bn\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"top_conv\\\", \\\"class_name\\\": \\\"Conv2D\\\", \\\"config\\\": {\\\"name\\\": \\\"top_conv\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"filters\\\": 1536, \\\"kernel_size\\\": [1, 1], \\\"strides\\\": [1, 1], \\\"padding\\\": \\\"same\\\", \\\"data_format\\\": \\\"channels_last\\\", \\\"dilation_rate\\\": [1, 1], \\\"activation\\\": \\\"linear\\\", \\\"use_bias\\\": false, \\\"kernel_initializer\\\": {\\\"class_name\\\": \\\"VarianceScaling\\\", \\\"config\\\": {\\\"scale\\\": 2.0, \\\"mode\\\": \\\"fan_out\\\", \\\"distribution\\\": \\\"normal\\\", \\\"seed\\\": null}}, \\\"bias_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"kernel_regularizer\\\": null, \\\"bias_regularizer\\\": null, \\\"activity_regularizer\\\": null, \\\"kernel_constraint\\\": null, \\\"bias_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"block7b_add\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"top_bn\\\", \\\"class_name\\\": \\\"BatchNormalization\\\", \\\"config\\\": {\\\"name\\\": \\\"top_bn\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"axis\\\": 3, \\\"momentum\\\": 0.99, \\\"epsilon\\\": 0.001, \\\"center\\\": true, \\\"scale\\\": true, \\\"beta_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"gamma_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"moving_mean_initializer\\\": {\\\"class_name\\\": \\\"Zeros\\\", \\\"config\\\": {}}, \\\"moving_variance_initializer\\\": {\\\"class_name\\\": \\\"Ones\\\", \\\"config\\\": {}}, \\\"beta_regularizer\\\": null, \\\"gamma_regularizer\\\": null, \\\"beta_constraint\\\": null, \\\"gamma_constraint\\\": null}, \\\"inbound_nodes\\\": [[[\\\"top_conv\\\", 0, 0, {}]]]}, {\\\"name\\\": \\\"top_activation\\\", \\\"class_name\\\": \\\"Activation\\\", \\\"config\\\": {\\\"name\\\": \\\"top_activation\\\", \\\"trainable\\\": true, \\\"dtype\\\": \\\"float32\\\", \\\"activation\\\": \\\"swish\\\"}, \\\"inbound_nodes\\\": [[[\\\"top_bn\\\", 0, 0, {}]]]}], \\\"input_layers\\\": [[\\\"input_5\\\", 0, 0]], \\\"output_layers\\\": [[\\\"top_activation\\\", 0, 0]]}, \\\"keras_version\\\": \\\"2.3.1\\\", \\\"backend\\\": \\\"tensorflow\\\"}\"\njson_string = json.loads(json.dumps(model_json))\nx_model = model_from_json(json_string)\n\n# Generalized mean pool - GeM\ngm_exp = tf.Variable(3.0, dtype = tf.float32)\ndef generalized_mean_pool_2d(X):\n    pool = (tf.reduce_mean(tf.abs(X**(gm_exp)),\n                        axis = [1, 2], \n                        keepdims = False) + 1.e-7)**(1./gm_exp)\n    return pool\n\n# UnFreeze all layers\nfor layer in x_model.layers:\n    layer.trainable = True\n    \n    # GeM\nlambda_layer = Lambda(generalized_mean_pool_2d)\nlambda_layer.trainable_weights.extend([gm_exp])\nx = lambda_layer(x_model.output)\n    \n    # multi output\ngrapheme_root = Dense(168, activation = 'softmax', name = 'root')(x)\nvowel_diacritic = Dense(11, activation = 'softmax', name = 'vowel')(x)\nconsonant_diacritic = Dense(7, activation = 'softmax', name = 'consonant')(x)\n\n    # model\nmodel = Model(inputs = x_model.input, outputs = [grapheme_root, vowel_diacritic, consonant_diacritic])\n\n# Compile Model\nmodel.compile(optimizer = Adam(lr = 0.00016),\n                loss = {'root': 'categorical_crossentropy',\n                        'vowel': 'categorical_crossentropy',\n                        'consonant': 'categorical_crossentropy'},\n                loss_weights = {'root': 0.50,        \n                                'vowel': 0.25,\n                                'consonant': 0.25},\n                metrics = {'root': ['accuracy', tf.keras.metrics.Recall()],\n                            'vowel': ['accuracy', tf.keras.metrics.Recall()],\n                            'consonant': ['accuracy', tf.keras.metrics.Recall()] })\n\nfrom keras.models import load_model\n#model = load_model('/kaggle/input/kagglebengaliaihandwrittengraphemeclassification/KaggleKernelEfficientNetB3/model_weights/Train1_model_59.h5')\nmodel.load_weights('/kaggle/input/bg-yhm68/Train2_model_68.h5')\n\n# Model Summary\nprint(model.summary())\n\n# Multi Label Stratified Split stuff...\nmsss = MultilabelStratifiedShuffleSplit(n_splits = EPOCHS, test_size = TEST_SIZE, random_state = SEED)\n\n# CustomReduceLRonPlateau function\nbest_val_loss = np.Inf\ndef CustomReduceLRonPlateau(model, history, epoch):\n    global best_val_loss\n    \n    # ReduceLR Constants\n    monitor = 'val_root_loss'\n    patience = 5\n    factor = 0.75\n    min_lr = 1e-5\n\n    # Get Current LR\n    current_lr = float(K.get_value(model.optimizer.lr))\n    \n    # Print Current Learning Rate\n    print('Current LR: {0}'.format(current_lr))\n\n    # Monitor Best Value\n    current_val_loss = history[monitor][-1]\n    if current_val_loss < best_val_loss:\n        best_val_loss = current_val_loss\n    print('Best Vall Loss: {0}'.format(best_val_loss))\n\n    # Track last values\n    if len(history[monitor]) >= patience:\n        last5 = history[monitor][-5:]\n        print('Last: {0}'.format(last5))\n        best_in_last = min(last5)\n        print('Min value in Last: {0}'.format(best_in_last))\n\n        # Determine correction\n        if best_val_loss < best_in_last:\n            new_lr = current_lr * factor\n            if new_lr < min_lr:\n                new_lr = min_lr\n            print('ReduceLRonPlateau setting learning rate to: {0}'.format(new_lr))\n            K.set_value(model.optimizer.lr, new_lr)\n\n# History Placeholder\nhistory = {}\n\n# Epoch Training Loop\nfor epoch, msss_splits in zip(range(0, EPOCHS), msss.split(X_train, Y_train)):\n    print('=========== EPOCH {}'.format(epoch+69))\n\n    # Get train and test index, shuffle train indexes.\n    train_idx = msss_splits[0]\n    valid_idx = msss_splits[1]\n    np.random.shuffle(train_idx)\n    print('Train Length: {0}   First 10 indices: {1}'.format(len(train_idx), train_idx[:10]))    \n    print('Valid Length: {0}    First 10 indices: {1}'.format(len(valid_idx), valid_idx[:10]))\n\n    transforms_train = albumentations.Compose([\n        GridMask(num_grid=3, rotate=15, p=1),\n    ])\n\n    # Create Data Generators for Train and Valid\n    data_generator_train = TrainDataGenerator(X_train, \n                                            Y_train,\n                                            train_idx, \n                                            BATCH_SIZE, \n                                            (HEIGHT_NEW, WIDTH_NEW, CHANNELS),\n                                            img_dir = TRAIN_DIR, transform=transforms_train)\n    data_generator_val = TrainDataGenerator(X_train, \n                                            Y_train,\n                                            valid_idx,\n                                            BATCH_SIZE, \n                                            (HEIGHT_NEW, WIDTH_NEW, CHANNELS),\n                                            img_dir = TRAIN_DIR)\n\n    TRAIN_STEPS = int(len(data_generator_train))\n    VALID_STEPS = int(len(data_generator_val))\n    print('Train Generator Size: {0}'.format(len(data_generator_train)))\n    print('Validation Generator Size: {0}'.format(len(data_generator_val)))\n    \n    model.fit_generator(generator = data_generator_train,\n                        validation_data = data_generator_val,\n                        steps_per_epoch = TRAIN_STEPS,\n                        validation_steps = VALID_STEPS,\n                        epochs = 1,\n                        callbacks = [ModelCheckpointFull(RUN_NAME + 'model_' + str(epoch+69) + '.h5')],\n                        verbose = 1)\n\n    # Set and Concat Training History\n    temp_history = model.history.history\n    if epoch == 0:\n        history = temp_history\n    else:\n        for k in temp_history: history[k] = history[k] + temp_history[k]\n\n    # Custom ReduceLRonPlateau\n    CustomReduceLRonPlateau(model, history, epoch)\n\n    # Cleanup\n    del data_generator_train, data_generator_val, train_idx, valid_idx\n    gc.collect()\n\n# Plot Training Summaries\nplot_summaries(history, PLOT_NAME1, PLOT_NAME2)\n\n# Create Predictions\nrow_ids, targets = [], []\nid = 0\n\n# Loop through parquet files\nfor i in range(4):\n    img_df = pd.read_parquet(os.path.join(DATA_DIR, 'test_image_data_'+str(i)+'.parquet'))\n    img_df = img_df.drop('image_id', axis = 1)\n    \n    # Loop through rows in parquet file\n    for index, row in img_df.iterrows():\n        img = resize_image(row.values, WIDTH, HEIGHT, WIDTH_NEW, HEIGHT_NEW)\n        img = np.stack((img,)*CHANNELS, axis=-1)\n        image = img.reshape(-1, HEIGHT_NEW, WIDTH_NEW, 3)\n        \n        # Predict\n        preds = model.predict(image, verbose = 1)\n        for k in range(3):\n            row_ids.append('Test_' + str(id) + '_' + tgt_cols[k])\n            targets.append(np.argmax(preds[k]))\n        id += 1\n\n# Create and Save Submission File\nsubmission = pd.DataFrame({'row_id': row_ids, 'target': targets}, columns = ['row_id', 'target'])\nsubmission.to_csv('submission.csv', index = False)\nprint(submission.head(25))","execution_count":null,"outputs":[]}],"metadata":{"kernelspec":{"name":"python3","display_name":"Python 3"},"language_info":{"codemirror_mode":{"name":"ipython","version":3},"file_extension":".py","mimetype":"text/x-python","name":"python","nbconvert_exporter":"python","pygments_lexer":"ipython3","version":"3.6.6"},"colab":{"name":"Keras EfficientNet B3 Training + Inference_YANG_training_v3.0.ipynb","provenance":[],"collapsed_sections":[],"toc_visible":true},"accelerator":"GPU"},"nbformat":4,"nbformat_minor":4}