{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.12","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceType":"competition","sourceId":30892,"databundleVersionId":2715462}],"dockerImageVersionId":30636,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"## Please change the Wandb API into yours","metadata":{}},{"cell_type":"code","source":"!pip install torchsummary\n!pip install torchgeometry","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:49:51.693593Z","iopub.execute_input":"2024-02-01T04:49:51.693905Z","iopub.status.idle":"2024-02-01T04:50:18.815548Z","shell.execute_reply.started":"2024-02-01T04:49:51.693878Z","shell.execute_reply":"2024-02-01T04:50:18.814588Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"from torchsummary import summary\nfrom torchgeometry.losses import one_hot\nimport os\nimport pandas as pd\nimport numpy as np\nfrom PIL import Image\nimport cv2\nimport time\nimport imageio\nimport matplotlib.pyplot as plt\nimport time\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.optim as optim\nfrom torch.optim import lr_scheduler\nfrom torch import Tensor\nfrom torch.utils.data import Dataset, DataLoader, random_split\nfrom torchvision.transforms import Resize, PILToTensor, ToPILImage, Compose, InterpolationMode\nfrom collections import OrderedDict\nimport wandb","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2024-02-01T04:50:18.817287Z","iopub.execute_input":"2024-02-01T04:50:18.817588Z","iopub.status.idle":"2024-02-01T04:50:26.589827Z","shell.execute_reply.started":"2024-02-01T04:50:18.817560Z","shell.execute_reply":"2024-02-01T04:50:26.588922Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"!nvidia-smi -L","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:26.590981Z","iopub.execute_input":"2024-02-01T04:50:26.591240Z","iopub.status.idle":"2024-02-01T04:50:27.562388Z","shell.execute_reply.started":"2024-02-01T04:50:26.591209Z","shell.execute_reply":"2024-02-01T04:50:27.561478Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\ndevice","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:27.564832Z","iopub.execute_input":"2024-02-01T04:50:27.565126Z","iopub.status.idle":"2024-02-01T04:50:27.620936Z","shell.execute_reply.started":"2024-02-01T04:50:27.565096Z","shell.execute_reply":"2024-02-01T04:50:27.619988Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"!pip install segmentation-models-pytorch","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:27.622261Z","iopub.execute_input":"2024-02-01T04:50:27.622768Z","iopub.status.idle":"2024-02-01T04:50:46.834307Z","shell.execute_reply.started":"2024-02-01T04:50:27.622732Z","shell.execute_reply":"2024-02-01T04:50:46.833315Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import segmentation_models_pytorch as smp\n\nmodel = smp.DeepLabV3Plus(\n    encoder_name=\"timm-regnetx_320\",        \n    encoder_weights=\"imagenet\",     \n    in_channels=3,                  \n    classes=3     \n)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:46.835700Z","iopub.execute_input":"2024-02-01T04:50:46.836032Z","iopub.status.idle":"2024-02-01T04:50:50.031124Z","shell.execute_reply.started":"2024-02-01T04:50:46.836001Z","shell.execute_reply":"2024-02-01T04:50:50.030348Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Parameters","metadata":{}},{"cell_type":"code","source":"\nnum_classes = 3\nepochs = 30\nlearning_rate = 1e-04\nbatch_size = 12\ndisplay_step = 50\n\ncheckpoint_path = '/kaggle/working/unet_model.pth'\npretrained_path = \"/kaggle/input/unet-checkpoint/unet_model.pth\"\n\nloss_epoch_array = []\ntrain_accuracy = []\ntest_accuracy = []\nvalid_accuracy = []","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.032208Z","iopub.execute_input":"2024-02-01T04:50:50.032492Z","iopub.status.idle":"2024-02-01T04:50:50.039017Z","shell.execute_reply.started":"2024-02-01T04:50:50.032467Z","shell.execute_reply":"2024-02-01T04:50:50.038176Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Dataloader","metadata":{}},{"cell_type":"code","source":"transform = Compose([Resize((512, 512), interpolation=InterpolationMode.BILINEAR),\n                     PILToTensor()])","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.040036Z","iopub.execute_input":"2024-02-01T04:50:50.040370Z","iopub.status.idle":"2024-02-01T04:50:50.052392Z","shell.execute_reply.started":"2024-02-01T04:50:50.040332Z","shell.execute_reply":"2024-02-01T04:50:50.051619Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class UNetDataClass(Dataset):\n    def __init__(self, images_path, masks_path, transform):\n        super(UNetDataClass, self).__init__()\n        \n        images_list = os.listdir(images_path)\n        masks_list = os.listdir(masks_path)\n        \n        images_list = [images_path + image_name for image_name in images_list]\n        masks_list = [masks_path + mask_name for mask_name in masks_list]\n        \n        self.images_list = images_list\n        self.masks_list = masks_list\n        self.transform = transform\n        \n    def __getitem__(self, index):\n        img_path = self.images_list[index]\n        mask_path = self.masks_list[index]\n        \n        # Open image and mask\n        data = Image.open(img_path)\n        label = Image.open(mask_path)\n        \n        # Normalize\n        data = self.transform(data) / 255\n        label = self.transform(label) / 255\n        \n        label = torch.where(label>0.65, 1.0, 0.0)\n        \n        label[2, :, :] = 0.0001\n        label = torch.argmax(label, 0).type(torch.int64)\n        \n        return data, label\n    \n    def __len__(self):\n        return len(self.images_list)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.053584Z","iopub.execute_input":"2024-02-01T04:50:50.053931Z","iopub.status.idle":"2024-02-01T04:50:50.065032Z","shell.execute_reply.started":"2024-02-01T04:50:50.053907Z","shell.execute_reply":"2024-02-01T04:50:50.064306Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"images_path = \"/kaggle/input/bkai-igh-neopolyp/train/train/\"\nmasks_path =  \"/kaggle/input/bkai-igh-neopolyp/train_gt/train_gt/\"","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.069117Z","iopub.execute_input":"2024-02-01T04:50:50.069387Z","iopub.status.idle":"2024-02-01T04:50:50.079099Z","shell.execute_reply.started":"2024-02-01T04:50:50.069364Z","shell.execute_reply":"2024-02-01T04:50:50.078266Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"unet_dataset = UNetDataClass(images_path, masks_path, transform)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.080317Z","iopub.execute_input":"2024-02-01T04:50:50.080662Z","iopub.status.idle":"2024-02-01T04:50:50.216148Z","shell.execute_reply.started":"2024-02-01T04:50:50.080630Z","shell.execute_reply":"2024-02-01T04:50:50.215496Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_size = 0.9\nvalid_size = 0.1","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.217149Z","iopub.execute_input":"2024-02-01T04:50:50.217401Z","iopub.status.idle":"2024-02-01T04:50:50.221215Z","shell.execute_reply.started":"2024-02-01T04:50:50.217378Z","shell.execute_reply":"2024-02-01T04:50:50.220339Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"torch.manual_seed(42)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.222241Z","iopub.execute_input":"2024-02-01T04:50:50.222489Z","iopub.status.idle":"2024-02-01T04:50:50.242624Z","shell.execute_reply.started":"2024-02-01T04:50:50.222467Z","shell.execute_reply":"2024-02-01T04:50:50.241917Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_set, valid_set = random_split(unet_dataset, \n                                    [int(train_size * len(unet_dataset)) , \n                                     int(valid_size * len(unet_dataset))])","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.243551Z","iopub.execute_input":"2024-02-01T04:50:50.243828Z","iopub.status.idle":"2024-02-01T04:50:50.256127Z","shell.execute_reply.started":"2024-02-01T04:50:50.243786Z","shell.execute_reply":"2024-02-01T04:50:50.255277Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_dataloader = DataLoader(train_set, batch_size=batch_size, shuffle=True)\nvalid_dataloader = DataLoader(valid_set, batch_size=batch_size, shuffle=True)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.257276Z","iopub.execute_input":"2024-02-01T04:50:50.257530Z","iopub.status.idle":"2024-02-01T04:50:50.262531Z","shell.execute_reply.started":"2024-02-01T04:50:50.257507Z","shell.execute_reply":"2024-02-01T04:50:50.261749Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Data Augmentation","metadata":{}},{"cell_type":"code","source":"from albumentations import (\n    Compose,\n    RandomRotate90,\n    Flip,\n    Transpose,\n    ElasticTransform,\n    GridDistortion,\n    OpticalDistortion,\n    RandomBrightnessContrast,\n    HorizontalFlip,\n    VerticalFlip,\n    RandomGamma,\n    RGBShift,\n)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:50.263573Z","iopub.execute_input":"2024-02-01T04:50:50.263950Z","iopub.status.idle":"2024-02-01T04:50:52.079368Z","shell.execute_reply.started":"2024-02-01T04:50:50.263912Z","shell.execute_reply":"2024-02-01T04:50:52.078595Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"augmentation = Compose([\n    HorizontalFlip(p=0.5),\n    VerticalFlip(p=0.5),\n    RandomGamma (gamma_limit=(70, 130), eps=None, always_apply=False, p=0.2),\n    RGBShift(p=0.3, r_shift_limit=10, g_shift_limit=10, b_shift_limit=10),\n])","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.080422Z","iopub.execute_input":"2024-02-01T04:50:52.080947Z","iopub.status.idle":"2024-02-01T04:50:52.086368Z","shell.execute_reply.started":"2024-02-01T04:50:52.080911Z","shell.execute_reply":"2024-02-01T04:50:52.085467Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class SegDataClass(Dataset):\n    def __init__(self, images_path, masks_path, transform=None, augmentation=None):\n        super(SegDataClass, self).__init__()\n        \n        images_list = os.listdir(images_path)\n        masks_list = os.listdir(masks_path)\n        \n        images_list = [os.path.join(images_path, image_name) for image_name in images_list]\n        masks_list = [os.path.join(masks_path, mask_name) for mask_name in masks_list]\n        \n        self.images_list = images_list\n        self.masks_list = masks_list\n        self.transform = transform\n        self.augmentation = augmentation\n        \n    def __getitem__(self, index):\n        img_path = self.images_list[index]\n        mask_path = self.masks_list[index]\n        \n        # Open image and mask\n        data = Image.open(img_path)\n        label = Image.open(mask_path)\n        \n        # Augmentation\n        if self.augmentation:\n            augmented = self.augmentation(image=np.array(data), mask=np.array(label))\n            data = Image.fromarray(augmented['image'])\n            label = Image.fromarray(augmented['mask'])\n        \n        # Normalize\n        data = self.transform(data) / 255\n        label = self.transform(label) / 255\n        \n        label = torch.where(label > 0.65, 1.0, 0.0)\n        label[2, :, :] = 0.0001\n        label = torch.argmax(label, 0).type(torch.int64)\n        \n        return data, label\n    \n    def __len__(self):\n        return len(self.images_list)\n\n\n# transform = transforms.ToTensor()\naug_dataset = SegDataClass(images_path, masks_path, transform=transform, augmentation=augmentation)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.087564Z","iopub.execute_input":"2024-02-01T04:50:52.087926Z","iopub.status.idle":"2024-02-01T04:50:52.119109Z","shell.execute_reply.started":"2024-02-01T04:50:52.087895Z","shell.execute_reply":"2024-02-01T04:50:52.118285Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"torch.manual_seed(42)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.120175Z","iopub.execute_input":"2024-02-01T04:50:52.121057Z","iopub.status.idle":"2024-02-01T04:50:52.127466Z","shell.execute_reply.started":"2024-02-01T04:50:52.121024Z","shell.execute_reply":"2024-02-01T04:50:52.126583Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_aug_set, valid_aug_set = random_split(aug_dataset, \n                                    [int(train_size * len(aug_dataset)) , \n                                     int(valid_size * len(aug_dataset))])","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.128541Z","iopub.execute_input":"2024-02-01T04:50:52.128855Z","iopub.status.idle":"2024-02-01T04:50:52.134703Z","shell.execute_reply.started":"2024-02-01T04:50:52.128820Z","shell.execute_reply":"2024-02-01T04:50:52.134052Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"from torch.utils.data import ConcatDataset\ntrain_dataloader = DataLoader(train_aug_set, batch_size=batch_size, shuffle=True)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.135881Z","iopub.execute_input":"2024-02-01T04:50:52.136110Z","iopub.status.idle":"2024-02-01T04:50:52.151814Z","shell.execute_reply.started":"2024-02-01T04:50:52.136089Z","shell.execute_reply":"2024-02-01T04:50:52.151015Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Model","metadata":{}},{"cell_type":"markdown","source":"**Residual Block**","metadata":{}},{"cell_type":"code","source":"class ResidualBlock(nn.Module):\n    def __init__(self, in_channels, out_channels):\n        super(ResidualBlock, self).__init__()\n\n        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1)\n        self.bn1 = nn.BatchNorm2d(out_channels)\n        self.relu = nn.ReLU(inplace=True)\n\n        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)\n        self.bn2 = nn.BatchNorm2d(out_channels)\n        \n        self.bn3 = nn.BatchNorm2d(out_channels)\n        \n        self.dropout = nn.Dropout(p=0.3)\n\n    def forward(self, x):\n        residual = x\n\n        out = self.conv1(x)\n        out = self.bn1(out)\n        out = self.relu(out)\n        \n        out = self.dropout(out)\n\n        out = self.conv2(out)\n        out = self.bn2(out)\n        \n        residual = self.conv1(residual)\n        residual = self.bn3(residual)\n\n        out += residual  \n        out = self.relu(out)\n\n        return out","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.153265Z","iopub.execute_input":"2024-02-01T04:50:52.153527Z","iopub.status.idle":"2024-02-01T04:50:52.163857Z","shell.execute_reply.started":"2024-02-01T04:50:52.153505Z","shell.execute_reply":"2024-02-01T04:50:52.163165Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"**Encoder Block**","metadata":{}},{"cell_type":"code","source":"class encoder_block(nn.Module):\n    def __init__(self, in_channels, out_channels):\n        super(encoder_block, self).__init__()\n        \n        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        \n        self.bn1 = nn.BatchNorm2d(out_channels)\n        self.bn2 = nn.BatchNorm2d(out_channels)\n        \n        self.relu = nn.ReLU()\n        self.dropout = nn.Dropout(p=0.3)\n        self.max_pool = nn.MaxPool2d(kernel_size=2, stride=2)\n        \n    def forward(self, x):\n        x = self.conv1(x)\n        x = self.bn1(x)\n        x = self.relu(x)\n        \n        x = self.dropout(x)\n        \n        x = self.conv2(x)\n        x = self.bn2(x)\n        x = self.relu(x)\n        \n        next_layer = self.max_pool(x)\n        skip_layer = x\n        \n        return next_layer, skip_layer","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.164833Z","iopub.execute_input":"2024-02-01T04:50:52.165138Z","iopub.status.idle":"2024-02-01T04:50:52.176127Z","shell.execute_reply.started":"2024-02-01T04:50:52.165108Z","shell.execute_reply":"2024-02-01T04:50:52.175457Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class res_encoder_block(nn.Module):\n    def __init__(self, in_channels, out_channels):\n        super(res_encoder_block, self).__init__()\n        \n        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        \n        self.bn1 = nn.BatchNorm2d(out_channels)\n        self.bn2 = nn.BatchNorm2d(out_channels)\n        self.bn3 = nn.BatchNorm2d(out_channels)\n        \n        self.relu = nn.ReLU()\n        self.dropout = nn.Dropout(p=0.3)\n        self.max_pool = nn.MaxPool2d(kernel_size=2, stride=2)\n        \n    def forward(self, x):\n        residual = x\n        \n        x = self.conv1(x)\n        x = self.bn1(x)\n        x = self.relu(x)\n        \n        x = self.dropout(x)\n        \n        x = self.conv2(x)\n        x = self.bn2(x)\n        \n        residual = self.conv1(residual)\n        residual = self.bn3(residual)\n        \n        \n        x += residual  \n        x = self.relu(x)\n        \n        next_layer = self.max_pool(x)\n        skip_layer = x\n        \n        return next_layer, skip_layer","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.177126Z","iopub.execute_input":"2024-02-01T04:50:52.177396Z","iopub.status.idle":"2024-02-01T04:50:52.193464Z","shell.execute_reply.started":"2024-02-01T04:50:52.177373Z","shell.execute_reply":"2024-02-01T04:50:52.192647Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"**Decoder block**","metadata":{}},{"cell_type":"code","source":"class decoder_block(nn.Module):\n    def __init__(self, in_channels, out_channels):\n        super(decoder_block, self).__init__()\n        \n        self.transpose_conv = nn.ConvTranspose2d(in_channels, out_channels, kernel_size=2, stride=2)\n        \n        self.conv1 = nn.Conv2d(2 * out_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        \n        self.bn1 = nn.BatchNorm2d(out_channels)\n        self.bn2 = nn.BatchNorm2d(out_channels)\n        \n        self.relu = nn.ReLU() \n        self.dropout = nn.Dropout(p=0.3)\n    \n    def forward(self, x, skip_layer):\n        x = self.transpose_conv(x)\n        x = torch.cat([x, skip_layer], axis=1)\n        \n        x = self.conv1(x)\n        x = self.bn1(x)\n        x = self.relu(x)\n        \n        x = self.dropout(x)\n        \n        x = self.conv2(x)\n        x = self.bn2(x)\n        x = self.relu(x)\n        \n        return x","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.194506Z","iopub.execute_input":"2024-02-01T04:50:52.194754Z","iopub.status.idle":"2024-02-01T04:50:52.209900Z","shell.execute_reply.started":"2024-02-01T04:50:52.194732Z","shell.execute_reply":"2024-02-01T04:50:52.209151Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class res_decoder_block(nn.Module):\n    def __init__(self, in_channels, out_channels):\n        super(res_decoder_block, self).__init__()\n        \n        self.transpose_conv = nn.ConvTranspose2d(in_channels, out_channels, kernel_size=2, stride=2)\n        \n        self.conv1 = nn.Conv2d(2 * out_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        \n        self.bn1 = nn.BatchNorm2d(out_channels)\n        self.bn2 = nn.BatchNorm2d(out_channels)\n        \n        self.bn3 = nn.BatchNorm2d(out_channels)\n        \n        self.relu = nn.ReLU() \n        self.dropout = nn.Dropout(p=0.3)\n    \n    def forward(self, x, skip_layer):\n        x = self.transpose_conv(x)\n        x = torch.cat([x, skip_layer], axis=1)\n        \n        residual = x\n        \n        x = self.conv1(x)\n        x = self.bn1(x)\n        x = self.relu(x)\n        \n        x = self.dropout(x)\n        \n        x = self.conv2(x)\n        x = self.bn2(x)\n        \n        residual = self.conv1(residual)\n        residual = self.bn3(residual)\n        \n        x += residual\n        x = self.relu(x)\n        \n        return x","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.210973Z","iopub.execute_input":"2024-02-01T04:50:52.211202Z","iopub.status.idle":"2024-02-01T04:50:52.227968Z","shell.execute_reply.started":"2024-02-01T04:50:52.211182Z","shell.execute_reply":"2024-02-01T04:50:52.227168Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"**Bottle neck**","metadata":{}},{"cell_type":"code","source":"class bottleneck_block(nn.Module):\n    def __init__(self, in_channels, out_channels):\n        super(bottleneck_block, self).__init__()\n        \n        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=1, padding='same')\n        \n        self.bn1 = nn.BatchNorm2d(out_channels)\n        self.bn2 = nn.BatchNorm2d(out_channels)\n        \n        self.relu = nn.ReLU()\n        self.dropout = nn.Dropout(p=0.3)\n        \n    def forward(self, x):\n        x = self.conv1(x)\n        x = self.bn1(x)\n        x = self.relu(x)\n        \n        x = self.dropout(x)\n        \n        x = self.conv2(x)\n        x = self.bn2(x)\n        x = self.relu(x)\n        \n        return x","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.229082Z","iopub.execute_input":"2024-02-01T04:50:52.229483Z","iopub.status.idle":"2024-02-01T04:50:52.243792Z","shell.execute_reply.started":"2024-02-01T04:50:52.229449Z","shell.execute_reply":"2024-02-01T04:50:52.243132Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"**Unet model**","metadata":{}},{"cell_type":"code","source":"# UNet model\nclass UNet(nn.Module):\n    def __init__(self, n_class=3):\n        super(UNet, self).__init__()\n        # Encoder blocks\n        self.enc1 = encoder_block(3, 64)\n        self.enc2 = encoder_block(64, 128)\n        self.enc3 = res_encoder_block(128, 256)\n        self.enc4 = encoder_block(256, 512)\n        \n        # Bottleneck block\n        self.bottleneck = ResidualBlock(512, 1024)\n        \n        # Decoder blocks\n        self.dec1 = decoder_block(1024, 512)\n        self.dec2 = res_decoder_block(512, 256)\n        self.dec3 = decoder_block(256, 128)\n        self.dec4 = decoder_block(128, 64)\n        \n        # 1x1 convolution\n        self.out = nn.Conv2d(64, n_class, kernel_size=1, padding='same')\n        \n    def forward(self, image):\n        n1, s1 = self.enc1(image)\n        n2, s2 = self.enc2(n1)\n        n3, s3 = self.enc3(n2)\n        n4, s4 = self.enc4(n3)\n        \n        n5 = self.bottleneck(n4)\n        \n        n6 = self.dec1(n5, s4)\n        n7 = self.dec2(n6, s3)\n        n8 = self.dec3(n7, s2)\n        n9 = self.dec4(n8, s1)\n        \n        output = self.out(n9)\n        \n        return output","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.249027Z","iopub.execute_input":"2024-02-01T04:50:52.249318Z","iopub.status.idle":"2024-02-01T04:50:52.259162Z","shell.execute_reply.started":"2024-02-01T04:50:52.249294Z","shell.execute_reply":"2024-02-01T04:50:52.258323Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Loss function","metadata":{}},{"cell_type":"code","source":"class CEDiceLoss(nn.Module):\n    def __init__(self, weights) -> None:\n        super(CEDiceLoss, self).__init__()\n        self.eps: float = 1e-6\n        self.weights: torch.Tensor = weights\n\n    def forward(\n            self,\n            input: torch.Tensor,\n            target: torch.Tensor) -> torch.Tensor:\n        if not torch.is_tensor(input):\n            raise TypeError(\"Input type is not a torch.Tensor. Got {}\"\n                            .format(type(input)))\n        if not len(input.shape) == 4:\n            raise ValueError(\"Invalid input shape, we expect BxNxHxW. Got: {}\"\n                             .format(input.shape))\n        if not input.shape[-2:] == target.shape[-2:]:\n            raise ValueError(\"input and target shapes must be the same. Got: {}\"\n                             .format(input.shape, input.shape))\n        if not input.device == target.device:\n            raise ValueError(\n                \"input and target must be in the same device. Got: {}\" .format(\n                    input.device, target.device))\n        if not self.weights.shape[1] == input.shape[1]:\n            raise ValueError(\"The number of weights must equal the number of classes\")\n        if not torch.sum(self.weights).item() == 1:\n            raise ValueError(\"The sum of all weights must equal 1\")\n            \n        # cross entropy loss\n        celoss = nn.CrossEntropyLoss(self.weights)(input, target)\n        \n        # compute softmax over the classes axis\n        input_soft = F.softmax(input, dim=1)\n\n        # create the labels one hot tensor\n        target_one_hot = one_hot(target, num_classes=input.shape[1],\n                                 device=input.device, dtype=input.dtype)\n\n        # compute the actual dice score\n        dims = (2, 3)\n        intersection = torch.sum(input_soft * target_one_hot, dims)\n        cardinality = torch.sum(input_soft + target_one_hot, dims)\n\n        dice_score = 2. * intersection / (cardinality + self.eps)\n        \n        dice_score = torch.sum(dice_score * self.weights, dim=1)\n        \n        return torch.mean(1. - dice_score) + celoss\n#         return dice_score","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.260292Z","iopub.execute_input":"2024-02-01T04:50:52.260611Z","iopub.status.idle":"2024-02-01T04:50:52.276963Z","shell.execute_reply.started":"2024-02-01T04:50:52.260580Z","shell.execute_reply":"2024-02-01T04:50:52.276228Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Training","metadata":{}},{"cell_type":"markdown","source":"**Initialize weights**","metadata":{}},{"cell_type":"code","source":"def weights_init(model):\n    if isinstance(model, nn.Linear):\n        # Xavier Distribution\n        torch.nn.init.xavier_uniform_(model.weight)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.277974Z","iopub.execute_input":"2024-02-01T04:50:52.278227Z","iopub.status.idle":"2024-02-01T04:50:52.295741Z","shell.execute_reply.started":"2024-02-01T04:50:52.278205Z","shell.execute_reply":"2024-02-01T04:50:52.294951Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"def save_model(model, optimizer, path):\n    checkpoint = {\n        \"model\": model.state_dict(),\n        \"optimizer\": optimizer.state_dict(),\n    }\n    torch.save(checkpoint, path)\n\ndef load_model(model, optimizer, path):\n    checkpoint = torch.load(path)\n    model.load_state_dict(checkpoint[\"model\"])\n    optimizer.load_state_dict(checkpoint['optimizer'])\n    return model, optimizer","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.296985Z","iopub.execute_input":"2024-02-01T04:50:52.297305Z","iopub.status.idle":"2024-02-01T04:50:52.306242Z","shell.execute_reply.started":"2024-02-01T04:50:52.297272Z","shell.execute_reply":"2024-02-01T04:50:52.305454Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"**Train model**","metadata":{}},{"cell_type":"code","source":"# Train function for each epoch\ndef train(train_dataloader, valid_dataloader,learing_rate_scheduler, epoch, display_step):\n    print(f\"Start epoch #{epoch+1}, learning rate for this epoch: {learing_rate_scheduler.get_last_lr()}\")\n    start_time = time.time()\n    train_loss_epoch = 0\n    test_loss_epoch = 0\n    last_loss = 999999999\n    model.train()\n    for i, (data,targets) in enumerate(train_dataloader):\n        \n        # Load data into GPU\n        data, targets = data.to(device), targets.to(device)\n\n        optimizer.zero_grad()\n        outputs = model(data)\n\n        # Backpropagation, compute gradients\n        loss = loss_function(outputs, targets.long())\n        loss.backward()\n\n        optimizer.step()\n        \n        # Save loss\n        train_loss_epoch += loss.item()\n        if (i+1) % display_step == 0:\n#             accuracy = float(test(test_loader))\n            print('Train Epoch: {} [{}/{} ({}%)]\\tLoss: {:.4f}'.format(\n                epoch + 1, (i+1) * len(data), len(train_dataloader.dataset), 100 * (i+1) * len(data) / len(train_dataloader.dataset), \n                loss.item()))\n                  \n    print(f\"Done epoch #{epoch+1}, time for this epoch: {time.time()-start_time}s\")\n    train_loss_epoch/= (i + 1)\n    \n    # Evaluate the validation set\n    model.eval()\n    with torch.no_grad():\n        for data, target in valid_dataloader:\n            data, target = data.to(device), target.to(device)\n            test_output = model(data)\n            test_loss = loss_function(test_output, target)\n            test_loss_epoch += test_loss.item()\n            \n    test_loss_epoch/= (i+1)\n    \n    return train_loss_epoch , test_loss_epoch","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.307371Z","iopub.execute_input":"2024-02-01T04:50:52.307641Z","iopub.status.idle":"2024-02-01T04:50:52.319544Z","shell.execute_reply.started":"2024-02-01T04:50:52.307617Z","shell.execute_reply":"2024-02-01T04:50:52.318799Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# Test function\ndef test(dataloader):\n    test_loss = 0\n    correct = 0\n    with torch.no_grad():\n        for i, (data, targets) in enumerate(dataloader):\n            data, targets = data.to(device), targets.to(device)\n            outputs = model(data)\n            _, pred = torch.max(outputs, 1)\n            test_loss += targets.size(0)\n            correct += torch.sum(pred == targets).item()\n    return 100.0 * correct / test_loss","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.320568Z","iopub.execute_input":"2024-02-01T04:50:52.320946Z","iopub.status.idle":"2024-02-01T04:50:52.335141Z","shell.execute_reply.started":"2024-02-01T04:50:52.320923Z","shell.execute_reply":"2024-02-01T04:50:52.334405Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# model = Unet(in_channels=3, num_classes = 3)\n\ntry:\n    checkpoint = torch.load(pretrained_path)\n\n    new_state_dict = OrderedDict()\n    for k, v in checkpoint['model'].items():\n        name = k[7:] # remove `module.`\n        new_state_dict[name] = v\n    # load params\n    model.load_state_dict(new_state_dict)\n    model = nn.DataParallel(model)\n    model.to(device)\nexcept:\n    model.apply(weights_init)\n    model = nn.DataParallel(model)\n    model.to(device)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.336239Z","iopub.execute_input":"2024-02-01T04:50:52.336512Z","iopub.status.idle":"2024-02-01T04:50:52.579743Z","shell.execute_reply.started":"2024-02-01T04:50:52.336490Z","shell.execute_reply":"2024-02-01T04:50:52.578995Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"weights = torch.Tensor([[0.4, 0.55, 0.05]]).cuda()\nloss_function = CEDiceLoss(weights)\n\n# Define the optimizer (Adam optimizer)\noptimizer = optim.Adam(params=model.parameters(), lr=learning_rate)\ntry: \n    optimizer.load_state_dict(checkpoint['optimizer'])\nexcept:\n    pass\n\n# Learning rate scheduler\nlearing_rate_scheduler = lr_scheduler.StepLR(optimizer, step_size=4, gamma=0.6)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.580826Z","iopub.execute_input":"2024-02-01T04:50:52.581107Z","iopub.status.idle":"2024-02-01T04:50:52.589789Z","shell.execute_reply.started":"2024-02-01T04:50:52.581082Z","shell.execute_reply":"2024-02-01T04:50:52.589134Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"save_model(model, optimizer, checkpoint_path)","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.591009Z","iopub.execute_input":"2024-02-01T04:50:52.591408Z","iopub.status.idle":"2024-02-01T04:50:52.746913Z","shell.execute_reply.started":"2024-02-01T04:50:52.591377Z","shell.execute_reply":"2024-02-01T04:50:52.746112Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"wandb.login(\n    # set the wandb project where this run will be logged\n#     project= \"PolypSegment\", \n    key = \"\",\n)\nwandb.init(\n    project = \"\"\n)\n# Training loop\ntrain_loss_array = []\ntest_loss_array = []\nlast_loss = 9999999999999\nfor epoch in range(epochs):\n    train_loss_epoch = 0\n    test_loss_epoch = 0\n    (train_loss_epoch, test_loss_epoch) = train(train_dataloader, \n                                              valid_dataloader, \n                                              learing_rate_scheduler, epoch, display_step)\n    \n    if test_loss_epoch < last_loss:\n        save_model(model, optimizer, checkpoint_path)\n        last_loss = test_loss_epoch\n        \n    learing_rate_scheduler.step()\n    train_loss_array.append(train_loss_epoch)\n    test_loss_array.append(test_loss_epoch)\n    wandb.log({\"Train loss\": train_loss_epoch, \"Valid loss\": test_loss_epoch})","metadata":{"execution":{"iopub.status.busy":"2024-02-01T04:50:52.748036Z","iopub.execute_input":"2024-02-01T04:50:52.748333Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Plot the learning cure","metadata":{}},{"cell_type":"markdown","source":"**Create submission**","metadata":{}},{"cell_type":"code","source":"class UNetTestDataClass(Dataset):\n    def __init__(self, images_path, transform):\n        super(UNetTestDataClass, self).__init__()\n        \n        images_list = os.listdir(images_path)\n        images_list = [images_path+i for i in images_list]\n        \n        self.images_list = images_list\n        self.transform = transform\n        \n    def __getitem__(self, index):\n        img_path = self.images_list[index]\n        data = Image.open(img_path)\n        h = data.size[1]\n        w = data.size[0]\n        data = self.transform(data) / 255        \n        return data, img_path, h, w\n    \n    def __len__(self):\n        return len(self.images_list)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"path = '/kaggle/input/bkai-igh-neopolyp/test/test/'\nunet_test_dataset = UNetTestDataClass(path, transform)\ntest_dataloader = DataLoader(unet_test_dataset, batch_size=8, shuffle=True)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"for i, (data, path, h, w) in enumerate(test_dataloader):\n    img = data\n    break","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"fig, arr = plt.subplots(5, 2, figsize=(16, 12))\narr[0][0].set_title('Image');\narr[0][1].set_title('Predict');\n\nmodel.eval()\nwith torch.no_grad():\n    predict = model(img)\n\nfor i in range(5):\n    arr[i][0].imshow(img[i].permute(1, 2, 0));\n    arr[i][1].imshow(F.one_hot(torch.argmax(predict[i], 0).cpu()).float())","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model.eval()\nif not os.path.isdir(\"/kaggle/working/predicted_masks\"):\n    os.mkdir(\"/kaggle/working/predicted_masks\")\nfor _, (img, path, H, W) in enumerate(test_dataloader):\n    a = path\n    b = img\n    h = H\n    w = W\n    \n    with torch.no_grad():\n        predicted_mask = model(b)\n    for i in range(len(a)):\n        image_id = a[i].split('/')[-1].split('.')[0]\n        filename = image_id + \".png\"\n        mask2img = Resize((h[i].item(), w[i].item()), interpolation=InterpolationMode.NEAREST)(ToPILImage()(F.one_hot(torch.argmax(predicted_mask[i], 0)).permute(2, 0, 1).float()))\n        mask2img.save(os.path.join(\"/kaggle/working/predicted_masks/\", filename))","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"def rle_to_string(runs):\n    return ' '.join(str(x) for x in runs)\n\ndef rle_encode_one_mask(mask):\n    pixels = mask.flatten()\n    pixels[pixels > 0] = 255\n    use_padding = False\n    if pixels[0] or pixels[-1]:\n        use_padding = True\n        pixel_padded = np.zeros([len(pixels) + 2], dtype=pixels.dtype)\n        pixel_padded[1:-1] = pixels\n        pixels = pixel_padded\n    \n    rle = np.where(pixels[1:] != pixels[:-1])[0] + 2\n    if use_padding:\n        rle = rle - 1\n    rle[1::2] = rle[1::2] - rle[:-1:2]\n    return rle_to_string(rle)\n\ndef mask2string(dir):\n    ## mask --> string\n    strings = []\n    ids = []\n    ws, hs = [[] for i in range(2)]\n    for image_id in os.listdir(dir):\n        id = image_id.split('.')[0]\n        path = os.path.join(dir, image_id)\n        print(path)\n        img = cv2.imread(path)[:,:,::-1]\n        h, w = img.shape[0], img.shape[1]\n        for channel in range(2):\n            ws.append(w)\n            hs.append(h)\n            ids.append(f'{id}_{channel}')\n            string = rle_encode_one_mask(img[:,:,channel])\n            strings.append(string)\n    r = {\n        'ids': ids,\n        'strings': strings,\n    }\n    return r\n\n\nMASK_DIR_PATH = '/kaggle/working/predicted_masks' # change this to the path to your output mask folder\ndir = MASK_DIR_PATH\nres = mask2string(dir)\ndf = pd.DataFrame(columns=['Id', 'Expected'])\ndf['Id'] = res['ids']\ndf['Expected'] = res['strings']\ndf.to_csv(r'output.csv', index=False)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"","metadata":{},"outputs":[],"execution_count":null}]}