{"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":"## summary\n\n* 2.5d segmentation\n * segmentation_models_pytorch \n * Unet\n* use only 6 slices\n* slide inference\n* add rotate TTA","metadata":{}},{"cell_type":"code","source":"import hashlib\nimport pandas as pd\ndata_dir = '/kaggle/input/vesuvius-challenge-ink-detection/test'\n\nskip_fake_test = True\n\n## code to skip test if commit run:\na_file = f'{data_dir}/a/mask.png'\nwith open(a_file,'rb') as f:\n hash_md5 = hashlib.md5(f.read()).hexdigest()\nis_skip_test = hash_md5 == '0b0fffdc0e88be226673846a143bb3e0'\nprint('is_skip_test:',is_skip_test) ## bu demek ki, test ici bos, daha dogrusu dummy test var.\n\nif skip_fake_test and is_skip_test:\n skipp = True\n print('skipping test')\n submit_df = pd.DataFrame({'Id': ['a', 'b'], 'Predicted':['1 2', '1 2']})\n submit_df.to_csv('submission.csv', index=False)\nelse: skipp = False\n \nprint(skipp)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:14:59.885251Z","iopub.execute_input":"2023-06-14T17:14:59.885637Z","iopub.status.idle":"2023-06-14T17:14:59.908116Z","shell.execute_reply.started":"2023-06-14T17:14:59.885602Z","shell.execute_reply":"2023-06-14T17:14:59.907160Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import torch\ntorch.__version__","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:14:59.909648Z","iopub.execute_input":"2023-06-14T17:14:59.910401Z","iopub.status.idle":"2023-06-14T17:15:01.509663Z","shell.execute_reply.started":"2023-06-14T17:14:59.910368Z","shell.execute_reply":"2023-06-14T17:15:01.508596Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from torch.utils.data import DataLoader\nfrom torch.cuda.amp import autocast, GradScaler\nimport sys\nimport time\nimport torch as tc\nimport random\nfrom torch.utils.data import DataLoader, Dataset\nimport torch\nimport albumentations as A\nfrom albumentations.pytorch import ToTensorV2\nimport numpy as np\n\nfrom tqdm.auto import tqdm\n\nimport torch\nimport torch.nn as nn\nfrom torch.optim import AdamW\n\nfrom torch.utils.data import DataLoader, Dataset\nimport matplotlib.pyplot as plt\nimport cv2,gc\nimport os,warnings\nimport pandas as pd","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:01.511224Z","iopub.execute_input":"2023-06-14T17:15:01.511995Z","iopub.status.idle":"2023-06-14T17:15:03.539135Z","shell.execute_reply.started":"2023-06-14T17:15:01.511958Z","shell.execute_reply":"2023-06-14T17:15:03.538166Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"try:\n import segmentation_models_pytorch as smp\nexcept: \n !pip install --no-index --find-links=\"/kaggle/input/segmentation-models-pytorch-offline-installation/\" segmentation-models-pytorch\n import segmentation_models_pytorch as smp","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:03.545738Z","iopub.execute_input":"2023-06-14T17:15:03.548221Z","iopub.status.idle":"2023-06-14T17:15:22.053078Z","shell.execute_reply.started":"2023-06-14T17:15:03.548185Z","shell.execute_reply":"2023-06-14T17:15:22.052052Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## config","metadata":{}},{"cell_type":"code","source":"import os\nimport albumentations as A\nfrom albumentations.pytorch import ToTensorV2\n\nclass CFG:\n # ============== comp exp name =============\n comp_name = 'vesuvius'\n\n # comp_dir_path = './'\n comp_dir_path = '/kaggle/input/'\n comp_folder_name = 'vesuvius-challenge-ink-detection'\n # comp_dataset_path = f'{comp_dir_path}datasets/{comp_folder_name}/'\n comp_dataset_path = f'{comp_dir_path}{comp_folder_name}/'\n \n exp_name = 'vesuvius_2d_slide_exp002'\n\n # ============== pred target =============\n target_size = 1\n TTA=True\n \n # ============== model cfg =============\n model_name = 'Unet'\n # backbone = 'efficientnet-b0'\n backbone = 'se_resnext50_32x4d'\n\n in_chans = 6 # 65\n # ============== training cfg =============\n size = 448\n tile_size = 448\n stride = tile_size // 2\n\n batch_size = 16 # 32\n use_amp = True\n\n scheduler = 'GradualWarmupSchedulerV2'\n # scheduler = 'CosineAnnealingLR'\n epochs = 15\n\n warmup_factor = 10\n lr = 1e-4 / warmup_factor\n\n # ============== fold =============\n valid_id = 2\n\n objective_cv = 'binary' # 'binary', 'multiclass', 'regression'\n metric_direction = 'maximize' # maximize, 'minimize'\n # metrics = 'dice_coef'\n\n # ============== fixed =============\n pretrained = True\n inf_weight = 'best' # 'best'\n\n min_lr = 1e-6\n weight_decay = 1e-6\n max_grad_norm = 1000\n\n print_freq = 50\n num_workers = 4\n\n seed = 42\n\n # ============== augmentation =============\n train_aug_list = [\n # A.RandomResizedCrop(\n # size, size, scale=(0.85, 1.0)),\n A.Resize(size, size),\n A.HorizontalFlip(p=0.5),\n A.VerticalFlip(p=0.5),\n A.RandomBrightnessContrast(p=0.75),\n A.ShiftScaleRotate(p=0.75),\n A.OneOf([\n A.GaussNoise(var_limit=[10, 50]),\n A.GaussianBlur(),\n A.MotionBlur(),\n ], p=0.4),\n A.GridDistortion(num_steps=5, distort_limit=0.3, p=0.5),\n A.CoarseDropout(max_holes=1, max_width=int(size * 0.3), max_height=int(size * 0.3), \n mask_fill_value=0, p=0.5),\n # A.Cutout(max_h_size=int(size * 0.6),\n # max_w_size=int(size * 0.6), num_holes=1, p=1.0),\n A.Normalize(\n mean= [0] * in_chans,\n std= [1] * in_chans\n ),\n ToTensorV2(transpose_mask=True),\n ]\n\n valid_aug_list = [\n A.Resize(size, size),\n A.Normalize(\n mean= [0] * in_chans,\n std= [1] * in_chans\n ),\n ToTensorV2(transpose_mask=True),\n ]\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.054992Z","iopub.execute_input":"2023-06-14T17:15:22.055412Z","iopub.status.idle":"2023-06-14T17:15:22.071480Z","shell.execute_reply.started":"2023-06-14T17:15:22.055372Z","shell.execute_reply":"2023-06-14T17:15:22.070251Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"IS_DEBUG = False\nmode = 'train' if IS_DEBUG else 'test'\nTH = 0.25","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.073097Z","iopub.execute_input":"2023-06-14T17:15:22.073481Z","iopub.status.idle":"2023-06-14T17:15:22.084766Z","shell.execute_reply.started":"2023-06-14T17:15:22.073447Z","shell.execute_reply":"2023-06-14T17:15:22.083691Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.087967Z","iopub.execute_input":"2023-06-14T17:15:22.088398Z","iopub.status.idle":"2023-06-14T17:15:22.119215Z","shell.execute_reply.started":"2023-06-14T17:15:22.088371Z","shell.execute_reply":"2023-06-14T17:15:22.118291Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## helper","metadata":{}},{"cell_type":"code","source":"# ref.: https://www.kaggle.com/stainsby/fast-tested-rle\ndef rle(img):\n '''\n img: numpy array, 1 - mask, 0 - background\n Returns run length as string formated\n '''\n pixels = img.flatten()\n # pixels = (pixels >= thr).astype(int)\n \n pixels = np.concatenate([[0], pixels, [0]])\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n runs[1::2] -= runs[::2]\n return ' '.join(str(x) for x in runs)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.120855Z","iopub.execute_input":"2023-06-14T17:15:22.122089Z","iopub.status.idle":"2023-06-14T17:15:22.133998Z","shell.execute_reply.started":"2023-06-14T17:15:22.122055Z","shell.execute_reply":"2023-06-14T17:15:22.133077Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## dataset","metadata":{}},{"cell_type":"code","source":"def read_image(fragment_id):\n images = []\n\n # idxs = range(65)\n mid = 65 // 2\n start = mid - CFG.in_chans // 2\n end = mid + CFG.in_chans // 2\n idxs = range(start, end)\n\n for i in tqdm(idxs):\n \n image = cv2.imread(CFG.comp_dataset_path + f\"{mode}/{fragment_id}/surface_volume/{i:02}.tif\", 0)\n\n pad0 = (CFG.tile_size - image.shape[0] % CFG.tile_size)\n pad1 = (CFG.tile_size - image.shape[1] % CFG.tile_size)\n\n image = np.pad(image, [(0, pad0), (0, pad1)], constant_values=0)\n\n images.append(image)\n images = np.stack(images, axis=2)\n \n return images","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.135649Z","iopub.execute_input":"2023-06-14T17:15:22.136074Z","iopub.status.idle":"2023-06-14T17:15:22.144780Z","shell.execute_reply.started":"2023-06-14T17:15:22.136042Z","shell.execute_reply":"2023-06-14T17:15:22.144064Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_transforms(data, cfg):\n if data == 'train':\n aug = A.Compose(cfg.train_aug_list)\n elif data == 'valid':\n aug = A.Compose(cfg.valid_aug_list)\n\n # print(aug)\n return aug\n\nclass CustomDataset(Dataset):\n def __init__(self, images, cfg, labels=None, transform=None):\n self.images = images\n self.cfg = cfg\n self.labels = labels\n self.transform = transform\n\n def __len__(self):\n # return len(self.xyxys)\n return len(self.images)\n\n def __getitem__(self, idx):\n # x1, y1, x2, y2 = self.xyxys[idx]\n image = self.images[idx]\n data = self.transform(image=image)\n image = data['image']\n return image\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.145783Z","iopub.execute_input":"2023-06-14T17:15:22.147337Z","iopub.status.idle":"2023-06-14T17:15:22.157792Z","shell.execute_reply.started":"2023-06-14T17:15:22.147304Z","shell.execute_reply":"2023-06-14T17:15:22.156931Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def make_test_dataset(fragment_id):\n test_images = read_image(fragment_id)\n \n x1_list = list(range(0, test_images.shape[1]-CFG.tile_size+1, CFG.stride))\n y1_list = list(range(0, test_images.shape[0]-CFG.tile_size+1, CFG.stride))\n \n test_images_list = []\n xyxys = []\n for y1 in y1_list:\n for x1 in x1_list:\n y2 = y1 + CFG.tile_size\n x2 = x1 + CFG.tile_size\n \n test_images_list.append(test_images[y1:y2, x1:x2])\n xyxys.append((x1, y1, x2, y2))\n xyxys = np.stack(xyxys)\n \n test_dataset = CustomDataset(test_images_list, CFG, transform=get_transforms(data='valid', cfg=CFG))\n \n test_loader = DataLoader(test_dataset,\n batch_size=CFG.batch_size,\n shuffle=False,\n num_workers=CFG.num_workers, pin_memory=True, drop_last=False)\n \n return test_loader, xyxys","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.161388Z","iopub.execute_input":"2023-06-14T17:15:22.161656Z","iopub.status.idle":"2023-06-14T17:15:22.170828Z","shell.execute_reply.started":"2023-06-14T17:15:22.161633Z","shell.execute_reply":"2023-06-14T17:15:22.169847Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## model","metadata":{}},{"cell_type":"code","source":"class CustomModel(nn.Module):\n def __init__(self, cfg, weight=None):\n super().__init__()\n self.cfg = cfg\n\n self.encoder = smp.Unet(\n encoder_name=cfg.backbone, \n encoder_weights=weight,\n in_channels=cfg.in_chans,\n classes=cfg.target_size,\n activation=None,\n )\n\n def forward(self, image):\n output = self.encoder(image)\n output = output.squeeze(-1)\n return output\n\ndef build_model(cfg, weight=\"imagenet\"):\n print('model_name', cfg.model_name)\n print('backbone', cfg.backbone)\n\n model = CustomModel(cfg, weight)\n return model\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.172340Z","iopub.execute_input":"2023-06-14T17:15:22.172727Z","iopub.status.idle":"2023-06-14T17:15:22.184158Z","shell.execute_reply.started":"2023-06-14T17:15:22.172689Z","shell.execute_reply":"2023-06-14T17:15:22.183254Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class EnsembleModel(nn.Module):\n def __init__(self):\n super().__init__()\n self.model = nn.ModuleList()\n for fold in [1, 2, 3]:\n _model = build_model(CFG, weight=None)\n #_model.to(device)\n\n model_path = f'/kaggle/input/vesuvius-models-public/{CFG.exp_name}/vesuvius-models/Unet_fold{fold}_best.pth'\n state = torch.load(model_path)['model']\n _model.load_state_dict(state)\n _model.eval()\n\n self.model.append(_model)\n \n def forward(self,x):\n output=[]\n for m in self.model:\n output.append(m(x))\n output=torch.stack(output,dim=0).mean(0)\n return output\n \n ","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.190845Z","iopub.execute_input":"2023-06-14T17:15:22.191203Z","iopub.status.idle":"2023-06-14T17:15:22.198879Z","shell.execute_reply.started":"2023-06-14T17:15:22.191168Z","shell.execute_reply":"2023-06-14T17:15:22.197429Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def TTA(x:tc.Tensor,model:nn.Module):\n #x.shape=(batch,c,h,w)\n if CFG.TTA:\n shape=x.shape\n x=[x,*[tc.rot90(x,k=i,dims=(-2,-1)) for i in range(1,4)]]\n x=tc.cat(x,dim=0)\n x=model(x)\n x=torch.sigmoid(x)\n x=x.reshape(4,shape[0],*shape[2:])\n x=[tc.rot90(x[i],k=-i,dims=(-2,-1)) for i in range(4)]\n x=tc.stack(x,dim=0)\n return x.mean(0)\n else :\n x=model(x)\n x=torch.sigmoid(x)\n return x","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.200649Z","iopub.execute_input":"2023-06-14T17:15:22.201189Z","iopub.status.idle":"2023-06-14T17:15:22.213087Z","shell.execute_reply.started":"2023-06-14T17:15:22.201158Z","shell.execute_reply":"2023-06-14T17:15:22.212171Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"if mode == 'test':\n fragment_ids = sorted(os.listdir(CFG.comp_dataset_path + mode))\nelse:\n fragment_ids = [3]","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.214677Z","iopub.execute_input":"2023-06-14T17:15:22.215053Z","iopub.status.idle":"2023-06-14T17:15:22.224182Z","shell.execute_reply.started":"2023-06-14T17:15:22.215021Z","shell.execute_reply":"2023-06-14T17:15:22.223267Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model = EnsembleModel()\n# model = nn.DataParallel(model) # nn.DataParallel(model, device_ids=[0, 1])\nmodel = model.cuda()","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:22.225690Z","iopub.execute_input":"2023-06-14T17:15:22.226098Z","iopub.status.idle":"2023-06-14T17:15:57.919085Z","shell.execute_reply.started":"2023-06-14T17:15:22.225996Z","shell.execute_reply":"2023-06-14T17:15:57.913248Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## main","metadata":{}},{"cell_type":"code","source":"results = []\nif not skipp:\n for fragment_id in fragment_ids:\n\n test_loader, xyxys = make_test_dataset(fragment_id)\n\n binary_mask = cv2.imread(CFG.comp_dataset_path + f\"{mode}/{fragment_id}/mask.png\", 0)\n binary_mask = (binary_mask / 255).astype(int)\n\n ori_h = binary_mask.shape[0]\n ori_w = binary_mask.shape[1]\n # mask = mask / 255\n\n pad0 = (CFG.tile_size - binary_mask.shape[0] % CFG.tile_size)\n pad1 = (CFG.tile_size - binary_mask.shape[1] % CFG.tile_size)\n\n binary_mask = np.pad(binary_mask, [(0, pad0), (0, pad1)], constant_values=0)\n\n mask_pred = np.zeros(binary_mask.shape)\n mask_count = np.zeros(binary_mask.shape)\n\n for step, (images) in tqdm(enumerate(test_loader), total=len(test_loader)):\n images = images.cuda()\n batch_size = images.size(0)\n\n with torch.no_grad():\n y_preds = TTA(images,model).cpu().numpy()\n\n start_idx = step*CFG.batch_size\n end_idx = start_idx + batch_size\n for i, (x1, y1, x2, y2) in enumerate(xyxys[start_idx:end_idx]):\n mask_pred[y1:y2, x1:x2] += y_preds[i].reshape(mask_pred[y1:y2, x1:x2].shape)\n mask_count[y1:y2, x1:x2] += np.ones((CFG.tile_size, CFG.tile_size))\n\n\n print(f'mask_count_min: {mask_count.min()}')\n mask_pred /= mask_count\n\n fig, axes = plt.subplots(1, 3, figsize=(15, 8))\n axes[0].imshow(mask_count)\n axes[1].imshow(mask_pred.copy())\n\n\n\n mask_pred = mask_pred[:ori_h, :ori_w]\n binary_mask = binary_mask[:ori_h, :ori_w]\n\n mask_pred = (mask_pred >= TH).astype(int)\n mask_pred *= binary_mask\n axes[2].imshow(mask_pred)\n plt.show()\n\n inklabels_rle = rle(mask_pred)\n\n results.append((fragment_id, inklabels_rle))\n\n\n del mask_pred, mask_count\n del test_loader\n\n gc.collect()\n torch.cuda.empty_cache()","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.920540Z","iopub.status.idle":"2023-06-14T17:15:57.921087Z","shell.execute_reply.started":"2023-06-14T17:15:57.920792Z","shell.execute_reply":"2023-06-14T17:15:57.920818Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## submission","metadata":{}},{"cell_type":"code","source":"if not skipp:\n sub = pd.DataFrame(results, columns=['Id', 'Predicted'])\n sample_sub = pd.read_csv(CFG.comp_dataset_path + 'sample_submission.csv')\n sample_sub = pd.merge(sample_sub[['Id']], sub, on='Id', how='left')\n sample_sub.to_csv(\"submission.csv\", index=False)\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.922787Z","iopub.status.idle":"2023-06-14T17:15:57.923275Z","shell.execute_reply.started":"2023-06-14T17:15:57.923040Z","shell.execute_reply":"2023-06-14T17:15:57.923062Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.924913Z","iopub.status.idle":"2023-06-14T17:15:57.925421Z","shell.execute_reply.started":"2023-06-14T17:15:57.925170Z","shell.execute_reply":"2023-06-14T17:15:57.925191Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!mkdir -p submission_ensemble","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.927068Z","iopub.status.idle":"2023-06-14T17:15:57.927546Z","shell.execute_reply.started":"2023-06-14T17:15:57.927306Z","shell.execute_reply":"2023-06-14T17:15:57.927328Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!cp submission.csv submission_ensemble/submission25.csv","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.934310Z","iopub.status.idle":"2023-06-14T17:15:57.935120Z","shell.execute_reply.started":"2023-06-14T17:15:57.934865Z","shell.execute_reply":"2023-06-14T17:15:57.934888Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import glob\nfor f in glob.glob('*'):\n if not f.startswith('submission_ensemble'):\n !rm -rf {f}","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.936522Z","iopub.status.idle":"2023-06-14T17:15:57.937305Z","shell.execute_reply.started":"2023-06-14T17:15:57.937045Z","shell.execute_reply":"2023-06-14T17:15:57.937067Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## birinci solution is over","metadata":{}},{"cell_type":"code","source":"%reset -f","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.938665Z","iopub.status.idle":"2023-06-14T17:15:57.939444Z","shell.execute_reply.started":"2023-06-14T17:15:57.939184Z","shell.execute_reply":"2023-06-14T17:15:57.939208Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import hashlib\nimport pandas as pd\ndata_dir = '/kaggle/input/vesuvius-challenge-ink-detection/test'\n\n\nskip_fake_test = True\n\n## code to skip test if commit run:\na_file = f'{data_dir}/a/mask.png'\nwith open(a_file,'rb') as f:\n hash_md5 = hashlib.md5(f.read()).hexdigest()\nis_skip_test = hash_md5 == '0b0fffdc0e88be226673846a143bb3e0'\nprint('is_skip_test:',is_skip_test) ## bu demek ki, test ici bos, daha dogrusu dummy test var.\n\nif skip_fake_test and is_skip_test:\n skipp = True\n print('skipping test')\n submit_df = pd.DataFrame({'Id': ['a', 'b'], 'Predicted':['1 2', '1 2']})\n submit_df.to_csv('submission.csv', index=False)\nelse: skipp = False\n \nprint(skipp)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.940843Z","iopub.status.idle":"2023-06-14T17:15:57.941661Z","shell.execute_reply.started":"2023-06-14T17:15:57.941396Z","shell.execute_reply":"2023-06-14T17:15:57.941421Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import segmentation_models_pytorch as smp","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.943056Z","iopub.status.idle":"2023-06-14T17:15:57.943824Z","shell.execute_reply.started":"2023-06-14T17:15:57.943580Z","shell.execute_reply":"2023-06-14T17:15:57.943604Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## 3d_resnet_size384_stride2_bs3","metadata":{}},{"cell_type":"code","source":"import os,cv2\nimport gc\nimport sys\nimport random\nfrom glob import glob\nimport numpy as np\nimport pandas as pd\nfrom tqdm import tqdm\nimport matplotlib.pyplot as plt\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom torch.cuda import amp\nfrom torch.utils.data import Dataset, DataLoader\n\nsys.path.append(\"/kaggle/input/resnet3d\")\nfrom resnet3d import generate_model\nimport torch as tc","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.945182Z","iopub.status.idle":"2023-06-14T17:15:57.945937Z","shell.execute_reply.started":"2023-06-14T17:15:57.945698Z","shell.execute_reply":"2023-06-14T17:15:57.945720Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class CFG:\n # ============== comp exp name =============\n comp_name = 'vesuvius'\n\n # comp_dir_path = './'\n comp_dir_path = '/kaggle/input/'\n comp_folder_name = 'vesuvius-challenge-ink-detection'\n # comp_dataset_path = f'{comp_dir_path}datasets/{comp_folder_name}/'\n comp_dataset_path = f'{comp_dir_path}{comp_folder_name}/'\n \n exp_name = 'vesuvius_2d_slide_exp002'\n\n # ============== pred target =============\n target_size = 1\n\n # ============== model cfg =============\n model_name = 'Unet'\n #backbone = 'efficientnet-b5'\n #backbone = 'mit_b5'\n backbone = 'resnet3d'\n #backbone = 'resnext50_32x4d'\n pretrained = True\n\n in_chans = 16 # 65\n load_chans=16\n # ============== training cfg =============\n prd_size=384\n stride = prd_size // 2\n\n batch_size = 3 # 32\n use_amp = True\n\n seed = 42\n num_workers=2","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.947391Z","iopub.status.idle":"2023-06-14T17:15:57.948154Z","shell.execute_reply.started":"2023-06-14T17:15:57.947894Z","shell.execute_reply":"2023-06-14T17:15:57.947917Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# ref.: https://www.kaggle.com/stainsby/fast-tested-rle\ndef rle(img):\n '''\n img: numpy array, 1 - mask, 0 - background\n Returns run length as string formated\n '''\n pixels = img.flatten()\n # pixels = (pixels >= thr).astype(int)\n \n pixels = np.concatenate([[0], pixels, [0]])\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n runs[1::2] -= runs[::2]\n return ' '.join(str(x) for x in runs)\ndef normalization(x:tc.Tensor)->tc.Tensor:\n \"\"\"input.shape=(batch,f1,f2,...)\"\"\"\n #[batch,f1,f2]->dim[1,2]\n dim=list(range(1,x.ndim))\n mean=x.mean(dim=dim,keepdim=True)\n std=x.std(dim=dim,keepdim=True)\n return (x-mean)/(std+1e-9)\n\ndef get_folder_size(folder_path):\n total_size = 0\n for dirpath, dirnames, filenames in os.walk(folder_path):\n for f in filenames:\n fp = os.path.join(dirpath, f)\n total_size += os.path.getsize(fp)\n return total_size","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.949540Z","iopub.status.idle":"2023-06-14T17:15:57.950303Z","shell.execute_reply.started":"2023-06-14T17:15:57.950039Z","shell.execute_reply":"2023-06-14T17:15:57.950062Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import cupy as cp\nxp = cp\n\ndelta_lookup = {\n \"xx\": xp.array([[1, -2, 1]], dtype=float),\n \"yy\": xp.array([[1], [-2], [1]], dtype=float),\n \"xy\": xp.array([[1, -1], [-1, 1]], dtype=float),\n}\n\ndef operate_derivative(img_shape, pair):\n assert len(img_shape) == 2\n delta = delta_lookup[pair]\n fft = xp.fft.fftn(delta, img_shape)\n return fft * xp.conj(fft)\n\ndef soft_threshold(vector, threshold):\n return xp.sign(vector) * xp.maximum(xp.abs(vector) - threshold, 0)\n\ndef back_diff(input_image, dim):\n assert dim in (0, 1)\n r, n = xp.shape(input_image)\n size = xp.array((r, n))\n position = xp.zeros(2, dtype=int)\n temp1 = xp.zeros((r+1, n+1), dtype=float)\n temp2 = xp.zeros((r+1, n+1), dtype=float)\n \n temp1[position[0]:size[0], position[1]:size[1]] = input_image\n temp2[position[0]:size[0], position[1]:size[1]] = input_image\n \n size[dim] += 1\n position[dim] += 1\n temp2[position[0]:size[0], position[1]:size[1]] = input_image\n temp1 -= temp2\n size[dim] -= 1\n return temp1[0:size[0], 0:size[1]]\n\ndef forward_diff(input_image, dim):\n assert dim in (0, 1)\n r, n = xp.shape(input_image)\n size = xp.array((r, n))\n position = xp.zeros(2, dtype=int)\n temp1 = xp.zeros((r+1, n+1), dtype=float)\n temp2 = xp.zeros((r+1, n+1), dtype=float)\n \n size[dim] += 1\n position[dim] += 1\n\n temp1[position[0]:size[0], position[1]:size[1]] = input_image\n temp2[position[0]:size[0], position[1]:size[1]] = input_image\n \n size[dim] -= 1\n temp2[0:size[0], 0:size[1]] = input_image\n temp1 -= temp2\n size[dim] += 1\n return -temp1[position[0]:size[0], position[1]:size[1]]\n\ndef iter_deriv(input_image, b, scale, mu, dim1, dim2):\n g = back_diff(forward_diff(input_image, dim1), dim2)\n d = soft_threshold(g + b, 1 / mu)\n b = b + (g - d)\n L = scale * back_diff(forward_diff(d - b, dim2), dim1)\n return L, b\n\ndef iter_xx(*args):\n return iter_deriv(*args, dim1=1, dim2=1)\n\ndef iter_yy(*args):\n return iter_deriv(*args, dim1=0, dim2=0)\n\ndef iter_xy(*args):\n return iter_deriv(*args, dim1=0, dim2=1)\n\ndef iter_sparse(input_image, bsparse, scale, mu):\n d = soft_threshold(input_image + bsparse, 1 / mu)\n bsparse = bsparse + (input_image - d)\n Lsparse = scale * (d - bsparse)\n return Lsparse, bsparse\n\ndef denoise_image(input_image, iter_num=100, fidelity=150, sparsity_scale=10, continuity_scale=0.5, mu=1):\n image_size = xp.shape(input_image)\n #print(\"Initialize denoising\")\n norm_array = (\n operate_derivative(image_size, \"xx\") + \n operate_derivative(image_size, \"yy\") + \n 2 * operate_derivative(image_size, \"xy\")\n )\n norm_array += (fidelity / mu) + sparsity_scale ** 2\n b_arrays = {\n \"xx\": xp.zeros(image_size, dtype=float),\n \"yy\": xp.zeros(image_size, dtype=float),\n \"xy\": xp.zeros(image_size, dtype=float),\n \"L1\": xp.zeros(image_size, dtype=float),\n }\n g_update = xp.multiply(fidelity / mu, input_image)\n for i in tqdm(range(iter_num), total=iter_num):\n #print(f\"Starting iteration {i+1}\")\n g_update = xp.fft.fftn(g_update)\n if i == 0:\n g = xp.fft.ifftn(g_update / (fidelity / mu)).real\n else:\n g = xp.fft.ifftn(xp.divide(g_update, norm_array)).real\n g_update = xp.multiply((fidelity / mu), input_image)\n \n #print(\"XX update\")\n L, b_arrays[\"xx\"] = iter_xx(g, b_arrays[\"xx\"], continuity_scale, mu)\n g_update += L\n \n #print(\"YY update\")\n L, b_arrays[\"yy\"] = iter_yy(g, b_arrays[\"yy\"], continuity_scale, mu)\n g_update += L\n \n #print(\"XY update\")\n L, b_arrays[\"xy\"] = iter_xy(g, b_arrays[\"xy\"], 2 * continuity_scale, mu)\n g_update += L\n \n #print(\"L1 update\")\n L, b_arrays[\"L1\"] = iter_sparse(g, b_arrays[\"L1\"], sparsity_scale, mu)\n g_update += L\n \n g_update = xp.fft.fftn(g_update)\n g = xp.fft.ifftn(xp.divide(g_update, norm_array)).real\n \n g[g < 0] = 0\n g -= g.min()\n g /= g.max()\n return g","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.951732Z","iopub.status.idle":"2023-06-14T17:15:57.952497Z","shell.execute_reply.started":"2023-06-14T17:15:57.952245Z","shell.execute_reply":"2023-06-14T17:15:57.952267Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def read_image(fragment_id):\n images = []\n\n # idxs = range(65)\n mid = 65 // 2\n start = mid - CFG.in_chans // 2\n end = mid + CFG.in_chans // 2\n idxs = range(start, end)\n\n for i in tqdm(idxs):\n image = cv2.imread(CFG.comp_dataset_path + f\"{mode}/{fragment_id}/surface_volume/{i:02}.tif\", 0)\n\n pad0 = (CFG.prd_size - image.shape[0] % CFG.prd_size)\n pad1 = (CFG.prd_size - image.shape[1] % CFG.prd_size)\n\n image = np.pad(image, [(0, pad0), (0, pad1)], constant_values=0)\n\n images.append(image)\n images = np.stack(images, axis=2)\n \n return images","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.953878Z","iopub.status.idle":"2023-06-14T17:15:57.954647Z","shell.execute_reply.started":"2023-06-14T17:15:57.954401Z","shell.execute_reply":"2023-06-14T17:15:57.954424Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class CustomDataset(Dataset):\n def __init__(self, images, cfg,xys, labels=None):\n self.images = images\n self.cfg = cfg\n self.labels = labels\n self.xys=xys\n\n def __len__(self):\n # return len(self.xyxys)\n return len(self.images)\n\n def __getitem__(self, idx):\n # x1, y1, x2, y2 = self.xyxys[idx]\n image = self.images[idx]\n image=tc.from_numpy(image).permute(2,0,1).to(tc.float32)/255\n image = (image - 0.45)/0.225\n return image,self.xys[idx]\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.955996Z","iopub.status.idle":"2023-06-14T17:15:57.956773Z","shell.execute_reply.started":"2023-06-14T17:15:57.956532Z","shell.execute_reply":"2023-06-14T17:15:57.956555Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def make_test_dataset(fragment_id):\n test_images = read_image(fragment_id)\n \n x1_list = list(range(0, test_images.shape[1]-CFG.prd_size+1, CFG.stride))\n y1_list = list(range(0, test_images.shape[0]-CFG.prd_size+1, CFG.stride))\n \n test_images_list = []\n xyxys = []\n for y1 in y1_list:\n for x1 in x1_list:\n y2 = y1 + CFG.prd_size\n x2 = x1 + CFG.prd_size\n if np.all(test_images[y1:y2, x1:x2]==0):\n continue\n test_images_list.append(test_images[y1:y2, x1:x2])\n xyxys.append((x1, y1, x2, y2))\n xyxys = np.stack(xyxys)\n \n test_dataset = CustomDataset(test_images_list, CFG,xys=xyxys)\n \n test_loader = DataLoader(test_dataset,\n batch_size=CFG.batch_size,\n shuffle=False,\n num_workers=CFG.num_workers, pin_memory=True, drop_last=False)\n \n return test_loader, xyxys","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.958160Z","iopub.status.idle":"2023-06-14T17:15:57.958918Z","shell.execute_reply.started":"2023-06-14T17:15:57.958673Z","shell.execute_reply":"2023-06-14T17:15:57.958695Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class Decoder(nn.Module):\n def __init__(self, encoder_dims, upscale):\n super().__init__()\n self.convs = nn.ModuleList([\n nn.Sequential(\n nn.Conv2d(encoder_dims[i]+encoder_dims[i-1], encoder_dims[i-1], 3, 1, 1, bias=False),\n nn.BatchNorm2d(encoder_dims[i-1]),\n nn.ReLU(inplace=True)\n ) for i in range(1, len(encoder_dims))])\n\n self.logit = nn.Conv2d(encoder_dims[0], 1, 1, 1, 0)\n self.up = nn.Upsample(scale_factor=upscale, mode=\"bilinear\")\n\n def forward(self, feature_maps):\n for i in range(len(feature_maps)-1, 0, -1):\n f_up = F.interpolate(feature_maps[i], scale_factor=2, mode=\"bilinear\")\n f = torch.cat([feature_maps[i-1], f_up], dim=1)\n f_down = self.convs[i-1](f)\n feature_maps[i-1] = f_down\n\n x = self.logit(feature_maps[0])\n mask = self.up(x)\n return mask\n \nclass SegModel(nn.Module):\n def __init__(self,model_depth=34):\n super().__init__()\n self.encoder = generate_model(model_depth=model_depth, n_input_channels=1)\n self.decoder = Decoder(encoder_dims=[64, 128, 256, 512], upscale=4)\n \n def forward(self, x):\n if x.ndim==4:\n x=x[:,None]\n \n feat_maps = self.encoder(x)\n feat_maps_pooled = [torch.mean(f, dim=2) for f in feat_maps]\n pred_mask = self.decoder(feat_maps_pooled)\n return pred_mask\n \nclass CustomModel(nn.Module):\n def __init__(self, cfg=CFG, weight=None):\n super().__init__()\n self.cfg = cfg\n\n if cfg.backbone==\"resnet3d\":\n self.encoder=SegModel()\n elif cfg.backbone[:3]!=\"mit\":\n self.encoder = smp.Unet(\n encoder_name=cfg.backbone, \n encoder_weights=weight,\n in_channels=cfg.in_chans,\n classes=cfg.target_size,\n activation=None,\n )\n else :\n self.encoder = smp.Unet(\n encoder_name=cfg.backbone, \n encoder_weights=weight,\n classes=cfg.target_size,\n activation=None,\n )\n print(self.encoder.encoder.patch_embed1.proj)\n out_channels=self.encoder.encoder.patch_embed1.proj.out_channels\n self.encoder.encoder.patch_embed1.proj=nn.Conv2d(cfg.in_chans,out_channels,7,4,3)\n\n def forward(self, images:torch.Tensor):\n #image.shape=(b,C,H,W)\n if images.ndim==4:\n images=images[:,None]\n images=normalization(images)\n output = self.encoder(images)\n return output\n\ndef build_model(cfg, weight=\"imagenet\"):\n print('model_name', cfg.model_name)\n print('backbone', cfg.backbone)\n\n model = CustomModel(cfg, weight)\n return model\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.960337Z","iopub.status.idle":"2023-06-14T17:15:57.961082Z","shell.execute_reply.started":"2023-06-14T17:15:57.960833Z","shell.execute_reply":"2023-06-14T17:15:57.960855Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def TTA(x:tc.Tensor,model:nn.Module):\n #x.shape=(batch,c,h,w)\n shape=x.shape\n x=[x,*[tc.rot90(x,k=i,dims=(-2,-1)) for i in range(1,4)]]\n x=tc.cat(x,dim=0)\n x=model(x)\n x=torch.sigmoid(x)\n x=x.reshape(4,shape[0],*shape[2:])\n x=[tc.rot90(x[i],k=-i,dims=(-2,-1)) for i in range(4)]\n x=tc.stack(x,dim=0)\n return x.mean(0)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.962464Z","iopub.status.idle":"2023-06-14T17:15:57.963243Z","shell.execute_reply.started":"2023-06-14T17:15:57.962967Z","shell.execute_reply":"2023-06-14T17:15:57.963017Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#in_submission=get_folder_size(\"/kaggle/input/vesuvius-challenge-ink-detection/test\")!=6732244267\nin_submission=True\nIS_DEBUG = False\nmode = 'train' if IS_DEBUG else 'test'\nTH = 0.55\nif mode == 'test':\n fragment_ids = sorted(os.listdir(CFG.comp_dataset_path + mode))\nelse:\n fragment_ids = [3]\n\nmodel = build_model(CFG)\nmodel.load_state_dict(tc.load(\"/kaggle/input/3d-resnet-baseline-inference-model-data/resnet3d-34_3d_seg_epoch_14.pth\"))\n# model = nn.DataParallel(model, device_ids=[0, 1])\nmodel = model.cuda()#.eval()\nmodel.training","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.964630Z","iopub.status.idle":"2023-06-14T17:15:57.965394Z","shell.execute_reply.started":"2023-06-14T17:15:57.965138Z","shell.execute_reply":"2023-06-14T17:15:57.965160Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"results = []\nif not skipp:\n for fragment_id in fragment_ids:\n if not in_submission:\n break\n test_loader, xyxys = make_test_dataset(fragment_id)\n\n binary_mask = cv2.imread(CFG.comp_dataset_path + f\"{mode}/{fragment_id}/mask.png\", 0)\n binary_mask = (binary_mask / 255).astype(int)\n\n ori_h = binary_mask.shape[0]\n ori_w = binary_mask.shape[1]\n\n pad0 = (CFG.prd_size - binary_mask.shape[0] % CFG.prd_size)\n pad1 = (CFG.prd_size - binary_mask.shape[1] % CFG.prd_size)\n\n binary_mask = np.pad(binary_mask, [(0, pad0), (0, pad1)], constant_values=0)\n\n mask_pred = np.zeros(binary_mask.shape)\n mask_count = np.zeros(binary_mask.shape)\n for step, (images,xys) in tqdm(enumerate(test_loader), total=len(test_loader)):\n images = images.cuda()\n batch_size = images.size(0)\n\n with torch.no_grad():\n y_preds=TTA(images,model)\n\n\n for k, (x1, y1, x2, y2) in enumerate(xys):\n mask_pred[y1:y2, x1:x2] += y_preds[k].squeeze(0).cpu().numpy()\n mask_count[y1:y2, x1:x2] += 1\n\n print(f'mask_count_min: {mask_count.min()}')\n mask_pred /= (mask_count+1e-7)\n\n\n fig, axes = plt.subplots(1, 4, figsize=(15, 8))\n axes[0].imshow(mask_count)\n axes[1].imshow(mask_pred.copy())\n\n mask_pred=xp.array(mask_pred)\n mask_pred=denoise_image(mask_pred, iter_num=250)\n mask_pred=mask_pred.get()\n axes[2].imshow(mask_pred)\n\n mask_pred = mask_pred[:ori_h, :ori_w]\n binary_mask = binary_mask[:ori_h, :ori_w]\n\n mask_pred = (mask_pred >= TH).astype(np.uint8)\n mask_pred=mask_pred.astype(int)\n mask_pred *= binary_mask\n\n axes[3].imshow(mask_pred)\n plt.show()\n\n inklabels_rle = rle(mask_pred)\n\n results.append((fragment_id, inklabels_rle))\n\n\n del mask_pred, mask_count\n del test_loader\n\n gc.collect()\n torch.cuda.empty_cache()\n plt.clf()\n fig.clear()\n plt.close(fig)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.966800Z","iopub.status.idle":"2023-06-14T17:15:57.967573Z","shell.execute_reply.started":"2023-06-14T17:15:57.967331Z","shell.execute_reply":"2023-06-14T17:15:57.967354Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"! cp /kaggle/input/vesuvius-challenge-ink-detection/sample_submission.csv submission.csv\nif not skipp:\n sub = pd.DataFrame(results, columns=['Id', 'Predicted'])\n #sub\n sample_sub = pd.read_csv(CFG.comp_dataset_path + 'sample_submission.csv')\n sample_sub = pd.merge(sample_sub[['Id']], sub, on='Id', how='left')\n #sample_sub\n sample_sub.to_csv(\"submission.csv\", index=False)\n print(\"ok\")","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.968939Z","iopub.status.idle":"2023-06-14T17:15:57.969703Z","shell.execute_reply.started":"2023-06-14T17:15:57.969466Z","shell.execute_reply":"2023-06-14T17:15:57.969488Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!cp submission.csv submission_ensemble/submission30.csv","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.971089Z","iopub.status.idle":"2023-06-14T17:15:57.971857Z","shell.execute_reply.started":"2023-06-14T17:15:57.971613Z","shell.execute_reply":"2023-06-14T17:15:57.971636Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"for f in glob('*'):\n if not f.startswith('submission_ensemble'):\n !rm -rf {f}","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.973229Z","iopub.status.idle":"2023-06-14T17:15:57.973992Z","shell.execute_reply.started":"2023-06-14T17:15:57.973746Z","shell.execute_reply":"2023-06-14T17:15:57.973768Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## inheritance inference","metadata":{}},{"cell_type":"code","source":"%reset -f","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.975381Z","iopub.status.idle":"2023-06-14T17:15:57.976143Z","shell.execute_reply.started":"2023-06-14T17:15:57.975888Z","shell.execute_reply":"2023-06-14T17:15:57.975911Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import hashlib\nimport pandas as pd\ndata_dir = '/kaggle/input/vesuvius-challenge-ink-detection/test'\n\n\nskip_fake_test = True\n\n## code to skip test if commit run:\na_file = f'{data_dir}/a/mask.png'\nwith open(a_file,'rb') as f:\n hash_md5 = hashlib.md5(f.read()).hexdigest()\nis_skip_test = hash_md5 == '0b0fffdc0e88be226673846a143bb3e0'\nprint('is_skip_test:',is_skip_test) ## bu demek ki, test ici bos, daha dogrusu dummy test var.\n\nif skip_fake_test and is_skip_test:\n skipp = True\n print('skipping test')\n submit_df = pd.DataFrame({'Id': ['a', 'b'], 'Predicted':['1 2', '1 2']})\n submit_df.to_csv('submission.csv', index=False)\nelse: skipp = False\n \nprint(skipp)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.977533Z","iopub.status.idle":"2023-06-14T17:15:57.978311Z","shell.execute_reply.started":"2023-06-14T17:15:57.978046Z","shell.execute_reply":"2023-06-14T17:15:57.978068Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\nimport os","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.979904Z","iopub.status.idle":"2023-06-14T17:15:57.980672Z","shell.execute_reply.started":"2023-06-14T17:15:57.980428Z","shell.execute_reply":"2023-06-14T17:15:57.980451Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import segmentation_models_pytorch as smp","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.982053Z","iopub.status.idle":"2023-06-14T17:15:57.982812Z","shell.execute_reply.started":"2023-06-14T17:15:57.982576Z","shell.execute_reply":"2023-06-14T17:15:57.982599Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import matplotlib.pyplot as plt\nimport math\nimport PIL.Image as Image\nimport torch\nfrom torch.utils.data import DataLoader\nimport torch.nn as nn\nfrom torch.optim import AdamW\nfrom torch.utils.data import DataLoader, Dataset, random_split\nimport torchvision.transforms as T\nimport cv2\nimport albumentations as A\nfrom albumentations.pytorch import ToTensorV2\nimport torch.optim as optim\nfrom torch.optim import lr_scheduler\nimport torch.backends.cudnn as cudnn\nimport torchvision\nimport time\nimport copy\nimport gc\nfrom tqdm.notebook import tqdm\nfrom tqdm import tqdm, trange\nfrom tqdm.auto import tqdm\n\ncudnn.benchmark = True\ndevice = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\nprint(torch.__version__)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.984228Z","iopub.status.idle":"2023-06-14T17:15:57.984980Z","shell.execute_reply.started":"2023-06-14T17:15:57.984741Z","shell.execute_reply":"2023-06-14T17:15:57.984764Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#config\nSUBMIT = True #if set to False, evaluate labels accuracy\nTILE_SIZE = 224\nSIZE = TILE_SIZE\nSTRIDE = TILE_SIZE// 2\n\n#specify scan images to load start / end (algorithm will find best positions automatically)\nSTART_SCAN = 28\nEND_SCAN = 36\nINPUT_CHANNELS = END_SCAN - START_SCAN + 1\n\nLR = 0.0001\nLOSS_FUNCTION = torch.nn.BCEWithLogitsLoss()\nTHRESHOLD = 0.45","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.986377Z","iopub.status.idle":"2023-06-14T17:15:57.987137Z","shell.execute_reply.started":"2023-06-14T17:15:57.986880Z","shell.execute_reply":"2023-06-14T17:15:57.986903Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def read_image(path):\n images = []\n\n slideids = range(START_SCAN, END_SCAN + 1)\n \n for i in slideids:\n #read images specified by start_scan and end_scan numbers\n image = cv2.imread(path + f\"/surface_volume/{i:02}.tif\", 0)\n \n pad0 = (TILE_SIZE- image.shape[0] % TILE_SIZE)\n pad1 = (TILE_SIZE- image.shape[1] % TILE_SIZE)\n \n #increase borders to match tile size\n image = np.pad(image, [(0, pad0), (0, pad1)], constant_values=0)\n \n images.append(image)\n del image\n gc.collect()\n #stack images (2D arrays) on axis 2\n images = np.stack(images, axis=2)\n \n return images\n ","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.988521Z","iopub.status.idle":"2023-06-14T17:15:57.989273Z","shell.execute_reply.started":"2023-06-14T17:15:57.989031Z","shell.execute_reply":"2023-06-14T17:15:57.989053Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class CustomDataset(Dataset):\n def __init__(self, images, transform = None):\n self.images = images\n self.transform = transform\n\n def __len__(self):\n return len(self.images)\n\n def __getitem__(self, idx):\n image = self.images[idx]\n if self.transform:\n data = self.transform(image = image)\n image = data['image']\n\n return image","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.990661Z","iopub.status.idle":"2023-06-14T17:15:57.991450Z","shell.execute_reply.started":"2023-06-14T17:15:57.991190Z","shell.execute_reply":"2023-06-14T17:15:57.991213Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def create_dataset(paths):\n \n images_list = []\n xyxys = []\n \n for path in paths:\n imageset = read_image(path)\n print(\"Input images - amount:\" + str(imageset.shape[2]) + \" height: \" + str(imageset.shape[0]) + \" width: \" + str(imageset.shape[1]))\n\n #split images into multiple smaller images (TILE_SIZE)\n x1_list = list(range(0, imageset.shape[1] - TILE_SIZE+ 1, TILE_SIZE))\n y1_list = list(range(0, imageset.shape[0] - TILE_SIZE+ 1, TILE_SIZE))\n\n for y1 in tqdm(y1_list,desc=\"Building dataset \"):\n for x1 in x1_list:\n y2 = y1 + TILE_SIZE\n x2 = x1 + TILE_SIZE\n images_list.append(imageset[y1:y2, x1:x2,:])\n xyxys.append((x1, y1, x2, y2)) \n \n xyxys = np.stack(xyxys)\n image_dataset = CustomDataset(images = images_list,transform=getTransforms())\n print(\"Dataset includes \" + str(len(images_list)) + \" - \" + str(TILE_SIZE) + \"x\" + str(TILE_SIZE) + \" images\")\n\n return image_dataset, xyxys","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.992830Z","iopub.status.idle":"2023-06-14T17:15:57.993598Z","shell.execute_reply.started":"2023-06-14T17:15:57.993355Z","shell.execute_reply":"2023-06-14T17:15:57.993377Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#trained different pretrained models and selected different architectures to build an average prediction\n#this function is used to create specified models and load the stored weights\ndef createModel(modelnr):\n if modelnr == 1: #Unet++ with better augmentation\n model = smp.UnetPlusPlus(\n encoder_name='resnext50_32x4d', \n encoder_weights=None, \n classes=1, \n in_channels=INPUT_CHANNELS,\n activation=None)\n optimizer = optim.Adam(model.parameters(), lr=LR) \n model = model.to(device)\n optimizer_ft = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)\n # Decay LR by a factor of 0.1 every 7 EPOCHS\n exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)\n model.load_state_dict(torch.load(\"/kaggle/input/trained-models/best-unetpp-resnext50_32x4d.pth\",map_location=torch.device(device)))\n \n elif modelnr == 2: #stride 3 Unet\n model = smp.Unet(\n encoder_name='resnext50_32x4d', \n encoder_weights=None, \n classes=1, \n in_channels=INPUT_CHANNELS,\n activation=None)\n optimizer = optim.Adam(model.parameters(), lr=LR) \n model = model.to(device)\n optimizer_ft = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)\n # Decay LR by a factor of 0.1 every 7 EPOCHS\n exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)\n model.load_state_dict(torch.load(\"/kaggle/input/trained-models/best-unet-resnext50_32x4d.pth\",map_location=torch.device(device)))\n \n elif modelnr == 3: #stride 3 Linknet\n model = smp.Linknet(\n encoder_name='resnet34', \n encoder_weights=None, \n classes=1, \n in_channels=INPUT_CHANNELS,\n activation=None)\n optimizer = optim.Adam(model.parameters(), lr=LR) \n model = model.to(device)\n optimizer_ft = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)\n # Decay LR by a factor of 0.1 every 7 EPOCHS\n exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)\n model.load_state_dict(torch.load(\"/kaggle/input/trained-models/best-linknet-resnet34.pth\",map_location=torch.device(device)))\n \n elif modelnr == 4: #stride 3 FPN\n model = smp.FPN(\n encoder_name='resnet34', \n encoder_weights=None, \n classes=1, \n in_channels=INPUT_CHANNELS,\n activation=None)\n optimizer = optim.Adam(model.parameters(), lr=LR) \n model = model.to(device)\n optimizer_ft = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)\n # Decay LR by a factor of 0.1 every 7 EPOCHS\n exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)\n model.load_state_dict(torch.load(\"/kaggle/input/trained-models/best-fpn-resnet34.pth\",map_location=torch.device(device)))\n \n return model ,optimizer","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.995108Z","iopub.status.idle":"2023-06-14T17:15:57.995875Z","shell.execute_reply.started":"2023-06-14T17:15:57.995632Z","shell.execute_reply":"2023-06-14T17:15:57.995655Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def getTransforms():\n #to apply normalization\n augList = [A.Normalize(mean= [0] * INPUT_CHANNELS,std= [1] * INPUT_CHANNELS),ToTensorV2(transpose_mask=True),]\n aug = A.Compose(augList)\n return aug","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.997271Z","iopub.status.idle":"2023-06-14T17:15:57.998039Z","shell.execute_reply.started":"2023-06-14T17:15:57.997786Z","shell.execute_reply":"2023-06-14T17:15:57.997809Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def predictTest(model,x):\n yMask = []\n #set model to evaluation mode\n model.eval()\n\n #turn off gradient tracking\n with torch.no_grad():\n for i in tqdm(range(len(x)), desc = \"Predict \" + str(model.name)):\n\n xPart = x[i]\n xPart = xPart.unsqueeze(0).to('cuda')\n yPred = model(xPart)\n \n predMask = yPred.squeeze()\n predMask = torch.sigmoid(predMask) \n\n #append predictions to list\n yMask.append(predMask.cpu().numpy())\n \n #convert list to array\n yMask = np.array(yMask)\n \n return yMask","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:57.999419Z","iopub.status.idle":"2023-06-14T17:15:58.000180Z","shell.execute_reply.started":"2023-06-14T17:15:57.999931Z","shell.execute_reply":"2023-06-14T17:15:57.999953Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Adapted from https://www.kaggle.com/code/stainsby/fast-tested-rle/notebook\n# and https://www.kaggle.com/code/kotaiizuka/faster-rle/notebook\ndef rle(output):\n #pixels = np.where(output.flatten().cpu() > THRESHOLD, 1, 0).astype(np.uint8)\n pixels = np.where(output.flatten() > THRESHOLD, 1, 0).astype(np.uint8)\n pixels[0] = 0\n pixels[-1] = 0\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 2\n runs[1::2] = runs[1::2] - runs[:-1:2]\n return ' '.join(str(x) for x in runs)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.001616Z","iopub.status.idle":"2023-06-14T17:15:58.002410Z","shell.execute_reply.started":"2023-06-14T17:15:58.002148Z","shell.execute_reply":"2023-06-14T17:15:58.002171Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#best accuracy models 6,7,8,9 with different architectures \n\nmodel1, _ = createModel(1) #unet++\nmodel2, _ = createModel(2) #unet\nmodel3, _ = createModel(3) #Linknet\nmodel4, _ = createModel(4) #FPN","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.003796Z","iopub.status.idle":"2023-06-14T17:15:58.004571Z","shell.execute_reply.started":"2023-06-14T17:15:58.004334Z","shell.execute_reply":"2023-06-14T17:15:58.004356Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def preparePlot(predMaskAvg, predMask1, predMask2, predMask3, predMask4, labelMask = None, threshold = 0.45):\n\n #initialize our figure\n cols = 6\n if labelMask is None:\n cols -= 1\n figure, ax = plt.subplots(nrows=1, ncols=cols, figsize=(10, 10))\n \n \n #plot \n ax[0].imshow((predMaskAvg > threshold) * 255)\n ax[1].imshow((predMask1 > threshold) * 255)\n ax[2].imshow((predMask2 > threshold) * 255)\n ax[3].imshow((predMask3 > threshold) * 255)\n ax[4].imshow((predMask4 > threshold) * 255)\n if labelMask is not None:\n ax[5].imshow((labelMask * 255).astype('int8'))\n \n #set the titles of the subplots\n ax[0].set_title(\"Pred Avg \" + str(threshold))\n ax[1].set_title(\"Unet++ \" + str(threshold))\n ax[2].set_title(\"Unet \" + str(threshold))\n ax[3].set_title(\"Linknet \" + str(threshold))\n ax[4].set_title(\"FPN \" + str(threshold))\n \n if labelMask is not None:\n ax[5].set_title(\"Label\")\n\n figure.tight_layout()\n \n figure.show()\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.005938Z","iopub.status.idle":"2023-06-14T17:15:58.006728Z","shell.execute_reply.started":"2023-06-14T17:15:58.006482Z","shell.execute_reply":"2023-06-14T17:15:58.006505Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def unpackImage(imageList,pad_h,pad_w):\n #assemble predicted image parts to one big image\n image = np.zeros(shape=(pad_h, pad_w))\n \n len_h = pad_h / TILE_SIZE\n len_w = pad_w / TILE_SIZE\n \n h = 0\n w = 0\n\n #fill predicted image with partial predictions\n for i in range(len(imageList)):\n image[h:h + TILE_SIZE,w:w + TILE_SIZE] = imageList[i,:,:]\n w += TILE_SIZE\n if w >= pad_w:\n h += TILE_SIZE\n w = 0\n \n return image","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.008144Z","iopub.status.idle":"2023-06-14T17:15:58.008906Z","shell.execute_reply.started":"2023-06-14T17:15:58.008667Z","shell.execute_reply":"2023-06-14T17:15:58.008690Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def predictFolder(path):\n startTime = time.time()\n #predict for submission\n TILE_SIZE= 224\n SIZE= TILE_SIZE\n \n #record accuracies\n acTreshold = []\n acAvg = []\n acUnetpp = []\n acUnet = []\n acLink = []\n acFdn = []\n \n #first testset\n paths = []\n paths.append(path)\n \n if os.path.isfile(path + \"/inklabels.png\"):\n labelExist = True\n labelPath = path + \"/inklabels.png\"\n label_mask = cv2.imread(labelPath, 0)\n labelImage = label_mask\n label_mask = (label_mask / 255).astype('float32')\n else:\n label_mask = None\n labelImage = label_mask\n labelExist = False\n\n test_set, _ = create_dataset(paths)\n\n yMask1 = predictTest(model1,test_set) #unet++\n yMask2 = predictTest(model2,test_set) #unet\n yMask3 = predictTest(model3,test_set) #linknet\n yMask4 = predictTest(model4,test_set) #fdn\n \n #build average prediction \n yMask = (yMask1 + yMask2 + yMask3 + yMask4) /4\n\n print(\"Predicted mask shape :\" + str(yMask.shape) )\n\n binary_mask = cv2.imread(path + \"/mask.png\", 0)\n binary_mask = (binary_mask / 255).astype('float32')\n\n #original size\n ori_h = binary_mask.shape[0]\n ori_w = binary_mask.shape[1]\n\n #padding again to fit with our tile size\n pad0 = (TILE_SIZE - binary_mask.shape[0] % TILE_SIZE)\n pad1 = (TILE_SIZE - binary_mask.shape[1] % TILE_SIZE)\n\n binary_mask_padded = np.pad(binary_mask, [(0, pad0), (0, pad1)], constant_values=0)\n\n pad_h = binary_mask_padded.shape[0]\n pad_w = binary_mask_padded.shape[1]\n \n #assemble predicted image parts to big image\n yPred = unpackImage(yMask,pad_h,pad_w)\n yPred1 = unpackImage(yMask1,pad_h,pad_w)\n yPred2 = unpackImage(yMask2,pad_h,pad_w)\n yPred3 = unpackImage(yMask3,pad_h,pad_w)\n yPred4 = unpackImage(yMask4,pad_h,pad_w)\n \n del yMask1\n del yMask2\n del yMask3\n del yMask4\n\n gc.collect()\n\n #back to original size\n yPred = yPred[:ori_h, :ori_w] \n yPred1 = yPred1[:ori_h, :ori_w] \n yPred2 = yPred2[:ori_h, :ori_w]\n yPred3 = yPred3[:ori_h, :ori_w] \n yPred4 = yPred4[:ori_h, :ori_w] \n\n #multiply with mask to set predictions to zero if corresponding mask is zero\n yPred = np.multiply(binary_mask, yPred)\n yPred1 = np.multiply(binary_mask, yPred1)\n yPred2 = np.multiply(binary_mask, yPred2)\n yPred3 = np.multiply(binary_mask, yPred3)\n yPred4 = np.multiply(binary_mask, yPred4)\n\n #show predictions \n preparePlot(yPred, yPred1, yPred2, yPred3, yPred4, labelImage, 0.5)\n\n\n \n #if labels exist, display accuracy for multiple thresholds\n if labelExist == True:\n for t in range(20,70,5):\n tact = t/100\n \n accAvg = ((((yPred > tact) * 1 == label_mask).astype('float32').sum())/label_mask.size) * 100\n acc1 = ((((yPred1 > tact) * 1 == label_mask).astype('float32').sum())/label_mask.size) * 100\n acc2 = ((((yPred2 > tact) * 1 == label_mask).astype('float32').sum())/label_mask.size) * 100\n acc3 = ((((yPred3 > tact) * 1 == label_mask).astype('float32').sum())/label_mask.size) * 100\n acc4 = ((((yPred4 > tact) * 1 == label_mask).astype('float32').sum())/label_mask.size) * 100\n \n acTreshold.append(tact)\n acAvg.append(accAvg)\n acUnetpp.append(acc1)\n acUnet.append(acc2)\n acLink.append(acc3)\n acFdn.append(acc4) \n \n #print(\"Average Accuracy: {:.2f} threshold: {:.2f}\".format(accAvg,tact))\n #print(\"Accuracy Unet: {:.2f} threshold: {:.2f}\".format(acc2,tact))\n #print(\"Accuracy Linknet: {:.2f} threshold: {:.2f}\".format(acc3,tact))\n #print(\"Accuracy FPN: {:.2f} threshold: {:.2f}\".format(acc4,tact))\n #print(\"*************************************************************\")\n \n #plot accuracy \n plt.figure(2,(15,15))\n plt.subplot(111)\n plt.plot(acTreshold, acAvg, label=\"Average\")\n plt.plot(acTreshold, acUnetpp, label=\"Unet++\")\n plt.plot(acTreshold, acUnet, label=\"Unet\")\n plt.plot(acTreshold, acLink, label=\"LinkNet\")\n plt.plot(acTreshold, acFdn, label=\"FDN\")\n\n plt.xlabel(\"Activation threshold\")\n plt.ylabel(\"Accuracy\")\n\n plt.title(\"Accuracy/Activation\")\n\n plt.legend()\n\n plt.show()\n \n #multiply with 1 to receive 0,1 values instead of True, False values\n yPred = (yPred > THRESHOLD) * 1\n yPred1 = (yPred1 > THRESHOLD) * 1\n yPred2 = (yPred2 > THRESHOLD) * 1\n yPred3 = (yPred3 > THRESHOLD) * 1\n yPred4 = (yPred4 > THRESHOLD) * 1\n\n #some statistics to compare data structure\n print(\"yPrediction:\" + \" min: \" + str(yPred.min()) + \" max: \" + str(yPred.max()) + \" shape: \" + str(yPred.shape) + \" size: \" + str(yPred.size))\n print(\"Binary Mask:\" + \" min: \" + str(binary_mask.min()) + \" max: \" + str(binary_mask.max()) + \" shape: \" + str(binary_mask.shape) + \" size: \" + str(binary_mask.size))\n if labelExist == True:\n print(\"Label Mask:\" + \" min: \" + str(label_mask.min()) + \" max: \" + str(label_mask.max()) + \" shape: \" + str(label_mask.shape) + \" size: \" + str(label_mask.size))\n print(\"**************************************************************************************************\")\n #use the average predictions for submission\n rleOutput = rle(yPred)\n\n endTime = time.time()\n totalTime = endTime - startTime\n print(\"Total time: {:.2f} seconds\".format(totalTime))\n \n\n \n return rleOutput","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.010393Z","iopub.status.idle":"2023-06-14T17:15:58.011142Z","shell.execute_reply.started":"2023-06-14T17:15:58.010886Z","shell.execute_reply":"2023-06-14T17:15:58.010908Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"if SUBMIT == False:\n _ = predictFolder(\"/kaggle/input/vesuvius-challenge-ink-detection/train/1\")","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.012518Z","iopub.status.idle":"2023-06-14T17:15:58.013277Z","shell.execute_reply.started":"2023-06-14T17:15:58.013039Z","shell.execute_reply":"2023-06-14T17:15:58.013061Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"if SUBMIT == False:\n _ = predictFolder(\"/kaggle/input/vesuvius-challenge-ink-detection/train/2\")","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.014646Z","iopub.status.idle":"2023-06-14T17:15:58.015419Z","shell.execute_reply.started":"2023-06-14T17:15:58.015153Z","shell.execute_reply":"2023-06-14T17:15:58.015176Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"if SUBMIT == False:\n _ = predictFolder(\"/kaggle/input/vesuvius-challenge-ink-detection/train/3\")","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.016786Z","iopub.status.idle":"2023-06-14T17:15:58.017564Z","shell.execute_reply.started":"2023-06-14T17:15:58.017324Z","shell.execute_reply":"2023-06-14T17:15:58.017347Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"if not skipp:\n rleOutput1 = predictFolder(\"/kaggle/input/vesuvius-challenge-ink-detection/test/a\")","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.018912Z","iopub.status.idle":"2023-06-14T17:15:58.019681Z","shell.execute_reply.started":"2023-06-14T17:15:58.019438Z","shell.execute_reply":"2023-06-14T17:15:58.019460Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"if not skipp:\n rleOutput2 = predictFolder(\"/kaggle/input/vesuvius-challenge-ink-detection/test/b\")","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.021061Z","iopub.status.idle":"2023-06-14T17:15:58.021827Z","shell.execute_reply.started":"2023-06-14T17:15:58.021586Z","shell.execute_reply":"2023-06-14T17:15:58.021608Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"if not skipp:\n results = []\n results.append(('a',rleOutput1))\n results.append(('b',rleOutput2))\n submission = pd.DataFrame(results, columns=['Id', 'Predicted'])\n submission.to_csv(\"submission.csv\", index=False)\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.023218Z","iopub.status.idle":"2023-06-14T17:15:58.023973Z","shell.execute_reply.started":"2023-06-14T17:15:58.023732Z","shell.execute_reply":"2023-06-14T17:15:58.023755Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!cp submission.csv submission_ensemble/submission_Inherit.csv","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.025392Z","iopub.status.idle":"2023-06-14T17:15:58.026151Z","shell.execute_reply.started":"2023-06-14T17:15:58.025898Z","shell.execute_reply":"2023-06-14T17:15:58.025920Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import glob\nfor f in glob.glob('*'):\n if not f.startswith('submission_ensemble'):\n !rm -rf {f}","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.027554Z","iopub.status.idle":"2023-06-14T17:15:58.028321Z","shell.execute_reply.started":"2023-06-14T17:15:58.028065Z","shell.execute_reply":"2023-06-14T17:15:58.028090Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## finalizing ","metadata":{}},{"cell_type":"code","source":"%reset -f","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.029654Z","iopub.status.idle":"2023-06-14T17:15:58.030428Z","shell.execute_reply.started":"2023-06-14T17:15:58.030168Z","shell.execute_reply":"2023-06-14T17:15:58.030190Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"## 4 submission oku ve decode","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.031783Z","iopub.status.idle":"2023-06-14T17:15:58.032557Z","shell.execute_reply.started":"2023-06-14T17:15:58.032313Z","shell.execute_reply":"2023-06-14T17:15:58.032337Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"## and this is for... the rle\nimport numba\n\n@numba.njit()\ndef rle_numba(pixels):\n size = len(pixels)\n points = []\n if pixels[0] == 1:\n flag = False\n points.append(1)\n else:\n flag = True\n for i in range(1, size):\n if pixels[i] != pixels[i-1]:\n if flag:\n points.append(i+1)\n flag = False\n else:\n points.append(i+1 - points[-1])\n flag = True\n if pixels[-1] == 1: points.append(size-points[-1]+1) \n return points\n\ndef rle_numba_encode(image):\n# pixels = image.flatten(order = 'F')\n pixels = image.flatten(order = 'C') ## bizden istedigi default olan C galiba .. bu comp icin\n\n points = rle_numba(pixels)\n return ' '.join(str(x) for x in points)\n\ndef rle_decode(mask_rle, shape=(384, 384)):\n '''\n mask_rle: run-length as string formated (start length)\n shape: (height,width) 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, order='C')","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.033928Z","iopub.status.idle":"2023-06-14T17:15:58.034697Z","shell.execute_reply.started":"2023-06-14T17:15:58.034455Z","shell.execute_reply":"2023-06-14T17:15:58.034477Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import os\nimport numpy as np\nimport pandas\nfrom PIL import Image\nimport pandas as pd","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.036104Z","iopub.status.idle":"2023-06-14T17:15:58.036866Z","shell.execute_reply.started":"2023-06-14T17:15:58.036624Z","shell.execute_reply":"2023-06-14T17:15:58.036648Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"## a ve b nin shapine gore is yapacaksin..\n## direk masklari alip bakabilirsin\ndata_dir = '/kaggle/input/vesuvius-challenge-ink-detection/test'\nrle = {}\n# weights = [1,1,1,1,1]\nweights = [1,1,1]\n\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.038241Z","iopub.status.idle":"2023-06-14T17:15:58.039022Z","shell.execute_reply.started":"2023-06-14T17:15:58.038762Z","shell.execute_reply":"2023-06-14T17:15:58.038784Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"mask_a = np.array(Image.open(os.path.join(f'{data_dir}/a', 'mask.png')))\nmask_b = np.array(Image.open(os.path.join(f'{data_dir}/b', 'mask.png')))\n\nheight_a,width_a = mask_a.shape\nheight_b,width_b = mask_b.shape","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.040414Z","iopub.status.idle":"2023-06-14T17:15:58.041175Z","shell.execute_reply.started":"2023-06-14T17:15:58.040917Z","shell.execute_reply":"2023-06-14T17:15:58.040941Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"height_a,width_a,height_b,width_b","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.042724Z","iopub.status.idle":"2023-06-14T17:15:58.043512Z","shell.execute_reply.started":"2023-06-14T17:15:58.043255Z","shell.execute_reply":"2023-06-14T17:15:58.043277Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# simdi submisionlari teker teker okuyaoim..","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.044875Z","iopub.status.idle":"2023-06-14T17:15:58.045652Z","shell.execute_reply.started":"2023-06-14T17:15:58.045410Z","shell.execute_reply":"2023-06-14T17:15:58.045433Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"rle_25 = pd.read_csv('/kaggle/working/submission_ensemble/submission25.csv')\nrle_30 = pd.read_csv('/kaggle/working/submission_ensemble/submission30.csv')\nrle_ii = pd.read_csv('/kaggle/working/submission_ensemble/submission_Inherit.csv')\n# rle_13 = pd.read_csv('/kaggle/working/submission_ensemble/submission_13Haz.csv')\n# rle_14 = pd.read_csv('/kaggle/working/submission_ensemble/submission_14Haz.csv')","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.047025Z","iopub.status.idle":"2023-06-14T17:15:58.047789Z","shell.execute_reply.started":"2023-06-14T17:15:58.047544Z","shell.execute_reply":"2023-06-14T17:15:58.047568Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# rle_25.Predicted[0]","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.049158Z","iopub.status.idle":"2023-06-14T17:15:58.049911Z","shell.execute_reply.started":"2023-06-14T17:15:58.049670Z","shell.execute_reply":"2023-06-14T17:15:58.049693Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"## now decode them.... adan baslayalim....\npreds_25 = rle_decode(rle_25.Predicted[0],(height_a,width_a))\npreds_30 = rle_decode(rle_30.Predicted[0],(height_a,width_a))\npreds_ii = rle_decode(rle_ii.Predicted[0],(height_a,width_a))\n# preds_13 = rle_decode(rle_13.Predicted[0],(height_a,width_a))\n# preds_14 = rle_decode(rle_14.Predicted[0],(height_a,width_a))\n\n## bunlari float yapip np.average with weights yapip sonra tekrar >.5 ile sonra orda rle ve\n## bu rle a olacak","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.051297Z","iopub.status.idle":"2023-06-14T17:15:58.052048Z","shell.execute_reply.started":"2023-06-14T17:15:58.051796Z","shell.execute_reply":"2023-06-14T17:15:58.051819Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# preds = np.stack([preds_25,preds_30,preds_ii,preds_13,preds_14]).astype(np.float32)\npreds = np.stack([preds_25,preds_30,preds_ii]).astype(np.float32)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.053423Z","iopub.status.idle":"2023-06-14T17:15:58.054182Z","shell.execute_reply.started":"2023-06-14T17:15:58.053931Z","shell.execute_reply":"2023-06-14T17:15:58.053954Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# np.sum(preds)\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.055616Z","iopub.status.idle":"2023-06-14T17:15:58.056637Z","shell.execute_reply.started":"2023-06-14T17:15:58.056395Z","shell.execute_reply":"2023-06-14T17:15:58.056418Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"## 25 30 inherit 13 ve 14 haziran modelleri sirasi ile weighted..\npreds = np.average(preds,axis=0,weights=weights)\npreds.shape","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.058022Z","iopub.status.idle":"2023-06-14T17:15:58.058783Z","shell.execute_reply.started":"2023-06-14T17:15:58.058541Z","shell.execute_reply":"2023-06-14T17:15:58.058563Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"preds = (preds>.5).astype(np.uint8)\nnp.sum(preds)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.060150Z","iopub.status.idle":"2023-06-14T17:15:58.060924Z","shell.execute_reply.started":"2023-06-14T17:15:58.060682Z","shell.execute_reply":"2023-06-14T17:15:58.060705Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"rle['a'] = rle_numba_encode(preds)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.062302Z","iopub.status.idle":"2023-06-14T17:15:58.063075Z","shell.execute_reply.started":"2023-06-14T17:15:58.062800Z","shell.execute_reply":"2023-06-14T17:15:58.062822Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"## simdi b icin:\n## now decode them.... adan baslayalim....\npreds_25 = rle_decode(rle_25.Predicted[1],(height_b,width_b))\npreds_30 = rle_decode(rle_30.Predicted[1],(height_b,width_b))\npreds_ii = rle_decode(rle_ii.Predicted[1],(height_b,width_b))\n# preds_13 = rle_decode(rle_13.Predicted[1],(height_b,width_b))\n# preds_14 = rle_decode(rle_14.Predicted[1],(height_b,width_b))\n\n# preds = np.stack([preds_25,preds_30,preds_ii,preds_13,preds_14]).astype(np.float32)\npreds = np.stack([preds_25,preds_30,preds_ii]).astype(np.float32)\n\n# mreds = preds.copy() ## burada anlirsan... averagelari da denersin ~~~~\npreds = np.average(preds,axis=0,weights=weights)\npreds = (preds>.5).astype(np.uint8)\nrle['b'] = rle_numba_encode(preds)\n\n## bunlari float yapip np.average with weights yapip sonra tekrar >.5 ile sonra orda rle ve\n## bu rle a olacak","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.064474Z","iopub.status.idle":"2023-06-14T17:15:58.065229Z","shell.execute_reply.started":"2023-06-14T17:15:58.064979Z","shell.execute_reply":"2023-06-14T17:15:58.065012Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import glob\nfor f in glob.glob('*'):\n !rm -rf {f}","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.066620Z","iopub.status.idle":"2023-06-14T17:15:58.067383Z","shell.execute_reply.started":"2023-06-14T17:15:58.067125Z","shell.execute_reply":"2023-06-14T17:15:58.067147Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"Id,Predicted\\na,\" + rle['a'] + \"\\nb,\" + rle['b'], file=open('submission.csv', 'w')) ","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.068723Z","iopub.status.idle":"2023-06-14T17:15:58.069506Z","shell.execute_reply.started":"2023-06-14T17:15:58.069253Z","shell.execute_reply":"2023-06-14T17:15:58.069275Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pd.read_csv('submission.csv')","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.070869Z","iopub.status.idle":"2023-06-14T17:15:58.071636Z","shell.execute_reply.started":"2023-06-14T17:15:58.071394Z","shell.execute_reply":"2023-06-14T17:15:58.071417Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Image.fromarray(preds*255)","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.072980Z","iopub.status.idle":"2023-06-14T17:15:58.073760Z","shell.execute_reply.started":"2023-06-14T17:15:58.073520Z","shell.execute_reply":"2023-06-14T17:15:58.073543Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# labels = np.array(Image.open('/kaggle/input/vesuvius-challenge-ink-detection/train/1/inklabels.png'))\n# labels = labels[-5454:]\n# labels.shape,np.unique(labels),preds.shape","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.075140Z","iopub.status.idle":"2023-06-14T17:15:58.075901Z","shell.execute_reply.started":"2023-06-14T17:15:58.075661Z","shell.execute_reply":"2023-06-14T17:15:58.075683Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# # # # NEDEN AYNISI OLMUHYOR ONA BAKMAK LAZIM.....\n# from sklearn.metrics import fbeta_score\n# fbeta_score(labels.flatten(), preds.flatten(), beta=0.5)\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.077483Z","iopub.status.idle":"2023-06-14T17:15:58.078367Z","shell.execute_reply.started":"2023-06-14T17:15:58.078096Z","shell.execute_reply":"2023-06-14T17:15:58.078120Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# tholds = [.05,.1,.15,.2,.25,.30,.35,.40,.45,.50,.55,.60,.65,.70, .75, .80]\n\n# for thi in tholds:\n# preds = mreds.copy()\n# # preds = mask*preds ## buna gerek kalmadi cunku mreds'i mask * preds sonrasina aldim...\n# preds = (preds>thi).astype(np.uint8)\n \n# score = fbeta_score(labels.flatten(), preds.flatten(), beta=0.5)\n# print(f'for thold of {thi},, the fbeta score is:{score}')\n\n\n# # for thold of 0.05,, the fbeta score is:0.4138839027303663\n# # for thold of 0.1,, the fbeta score is:0.46358726046431376\n# # for thold of 0.15,, the fbeta score is:0.501133748144812\n# # for thold of 0.2,, the fbeta score is:0.5320235519067721\n# # for thold of 0.25,, the fbeta score is:0.5604565047334683\n# # for thold of 0.3,, the fbeta score is:0.5691042296051885\n# # for thold of 0.35,, the fbeta score is:0.5719583491913619\n# # for thold of 0.4,, the fbeta score is:0.5692280584683906\n# # for thold of 0.45,, the fbeta score is:0.560449774441391\n# # for thold of 0.5,, the fbeta score is:0.5512158841316568\n# # for thold of 0.55,, the fbeta score is:0.5339695495590538\n# # for thold of 0.6,, the fbeta score is:0.5108621823429387\n# # for thold of 0.65,, the fbeta score is:0.48228067416494846\n# # for thold of 0.7,, the fbeta score is:0.44896378975670603\n# # for thold of 0.75,, the fbeta score is:0.4075783422540912\n# # for thold of 0.8,, the fbeta score is:0.3617596448857491","metadata":{"execution":{"iopub.status.busy":"2023-06-14T17:15:58.079783Z","iopub.status.idle":"2023-06-14T17:15:58.080553Z","shell.execute_reply.started":"2023-06-14T17:15:58.080314Z","shell.execute_reply":"2023-06-14T17:15:58.080339Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}