{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"This data set contains only 351 data. So using transfer learning is good intial point to tune the pretrained model.\nHere, MobileNet_V2 is used as model to tune and predict the segments.\nCode is extracted from https://www.tensorflow.org/tutorials/images/segmentation","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2022-08-12T12:31:55.094375Z","iopub.execute_input":"2022-08-12T12:31:55.095005Z","iopub.status.idle":"2022-08-12T12:31:55.130647Z","shell.execute_reply.started":"2022-08-12T12:31:55.094927Z","shell.execute_reply":"2022-08-12T12:31:55.12944Z"}}},{"cell_type":"code","source":"!pip install imutils\nimport tensorflow as tf\nfrom tensorflow.keras.layers import Conv2D, BatchNormalization, Activation, MaxPool2D, Conv2DTranspose, Concatenate, Input\nfrom tensorflow.keras.models import Model\nfrom tensorflow.keras.applications import MobileNetV2\n\nfrom tensorflow import keras\nimport numpy as np\nfrom tensorflow.keras.preprocessing.image import load_img\nimport random\nfrom sklearn.model_selection import train_test_split\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport imutils\n\nimg_size = (512, 512)\nnum_classes = 2\nbatch_size = 4","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class CustomeDataloader(keras.utils.Sequence):\n    \"\"\"Helper to iterate over the data (as Numpy arrays).\"\"\"\n\n    def __init__(self, batch_size, img_size, csv_data, root_dir):\n        self.batch_size = batch_size\n        self.img_size = img_size\n        self.csv_data = csv_data\n        self.root_dir = root_dir\n\n    def __len__(self):\n        return len(self.csv_data) // self.batch_size\n    \n    def rle2mask(self, mask_rle, shape=(1600,256)):\n        '''\n        mask_rle: run-length as string formated (start length)\n        shape: (width,height) of array to return \n        Returns numpy array, 1 - mask, 0 - background\n    \n        '''\n        s = mask_rle.split()\n        starts, lengths = [np.asarray(x, dtype=int) for x in (s[0:][::2], s[1:][::2])]\n        starts -= 1\n        ends = starts + lengths\n        img = np.zeros(shape[0]*shape[1], dtype=np.uint8)\n        for lo, hi in zip(starts, ends):\n            img[lo:hi] = 1\n        return img.reshape(shape).T\n\n    def __getitem__(self, idx):\n        \"\"\"Returns tuple (input, target) correspond to batch #idx.\"\"\"\n        i = idx * self.batch_size\n        batch_csv_data = self.csv_data.iloc[i : i + self.batch_size, :].values\n        x = np.zeros((self.batch_size,) + self.img_size + (3,), dtype=\"float32\")\n        y = np.zeros((self.batch_size,) + self.img_size, dtype=\"uint8\")\n        for j, path in enumerate(batch_csv_data):\n            path_img = self.root_dir + str(path[0]) + \".tiff\"\n            img = load_img(path_img)\n            img = tf.image.per_image_standardization(img)\n            x[j] = imutils.resize(np.array(img) ,height = 512, width=512) \n            img_mask = self.rle2mask(path[7], (path[4], path[3]))\n            y[j] = imutils.resize(img_mask ,height = 512, width=512) \n        return x,y","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Instantiate data Sequences for each split\ndata = pd.read_csv(\"/kaggle/input/hubmap-organ-segmentation/train.csv\")\nX_train, X_test, y_train, y_test = train_test_split(lung_data, lung_data, test_size=0.33, random_state=42)\ntrain_gen = CustomeDataloader(batch_size, img_size, X_train, \"/kaggle/input/hubmap-organ-segmentation/train_images/\")\nval_gen = CustomeDataloader(batch_size, img_size, X_test, \"/kaggle/input/hubmap-organ-segmentation/train_images/\")","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def conv_block(inputs, num_filters):\n    x = Conv2D(num_filters, 3, padding=\"same\")(inputs)\n    x = BatchNormalization()(x)\n    x = Activation(\"relu\")(x)\n\n    x = Conv2D(num_filters, 3, padding=\"same\")(x)\n    x = BatchNormalization()(x)\n    x = Activation(\"relu\")(x)\n\n    return x\n\ndef decoder_block(inputs, skip, num_filters):\n    x = Conv2DTranspose(num_filters, (2, 2), strides=2, padding=\"same\")(inputs)\n    x = Concatenate()([x, skip])\n    x = conv_block(x, num_filters)\n\n    return x\n\ndef build_mobilenetv2_unet(input_shape):    ## (512, 512, 3)\n    \"\"\" Input \"\"\"\n    inputs = Input(shape=input_shape)\n\n    \"\"\" Pre-trained MobileNetV2 \"\"\"\n    encoder = MobileNetV2(include_top=False, weights=\"imagenet\",\n        input_tensor=inputs, alpha=1.4)\n\n    \"\"\" Encoder \"\"\"   \n    s1 = encoder.get_layer(encoder.get_layer(index=0).name).output                ## (512 x 512)\n    s2 = encoder.get_layer(encoder.get_layer(index=11).name).output    ## (256 x 256)\n    s3 = encoder.get_layer(encoder.get_layer(index=29).name).output    ## (128 x 128)\n    s4 = encoder.get_layer(encoder.get_layer(index=56).name).output    ## (64 x 64)\n\n    \"\"\" Bridge \"\"\"\n    b1 = encoder.get_layer(\"block_13_expand_relu\").output   ## (32 x 32)\n\n    \"\"\" Decoder \"\"\"\n    d1 = decoder_block(b1, s4, 512)                         ## (64 x 64)\n    d2 = decoder_block(d1, s3, 256)                         ## (128 x 128)\n    d3 = decoder_block(d2, s2, 128)                         ## (256 x 256)\n    d4 = decoder_block(d3, s1, 64)                          ## (512 x 512)\n\n    \"\"\" Output \"\"\"\n    outputs = Conv2D(1, 1, padding=\"same\", activation=\"sigmoid\")(d4)\n\n    model = Model(inputs, outputs, name=\"MobileNetV2_U-Net\")\n    return model","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model = build_mobilenetv2_unet((512, 512, 3))\n\nmodel.compile(optimizer='adam',\n              loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),\n              metrics=[tf.keras.metrics.BinaryAccuracy(),\n                       tf.keras.metrics.FalseNegatives()])\n\nEPOCHS = 20\n\nmodel_history = model.fit(train_gen, epochs=EPOCHS,validation_data=val_gen)","metadata":{},"execution_count":null,"outputs":[]}]}