{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.11.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"nvidiaTeslaT4","dataSources":[{"sourceId":97984,"databundleVersionId":14096757,"sourceType":"competition"},{"sourceId":13746387,"sourceType":"datasetVersion","datasetId":8747012},{"sourceId":13816899,"sourceType":"datasetVersion","datasetId":8620533},{"sourceId":271051632,"sourceType":"kernelVersion"},{"sourceId":677607,"sourceType":"modelInstanceVersion","modelInstanceId":513841,"modelId":528480}],"dockerImageVersionId":31153,"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"
\n\n
\n\n

\n Import libraries\n

","metadata":{}},{"cell_type":"code","source":"import os\nimport cv2\nimport torch\nimport numpy as np\nimport pandas as pd\nimport torch.nn as nn\nimport albumentations as A\nimport torch.optim as optim\n\nfrom scipy import signal as scipy_signal\nfrom albumentations.pytorch import ToTensorV2\nfrom torch.utils.data import Dataset, DataLoader","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:15:06.710994Z","iopub.execute_input":"2025-12-22T14:15:06.711264Z","iopub.status.idle":"2025-12-22T14:15:11.270010Z","shell.execute_reply.started":"2025-12-22T14:15:06.711237Z","shell.execute_reply":"2025-12-22T14:15:11.269237Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n

\n Config(u can change parameters)\n

","metadata":{}},{"cell_type":"code","source":"class Config:\n batch_size = 10\n epochs = 1\n lr = 1e-4\n num_workers = 2\n img_size = (1240, 1024)\n target_length = 5000\n\nconfig = Config()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:15:11.271116Z","iopub.execute_input":"2025-12-22T14:15:11.271512Z","iopub.status.idle":"2025-12-22T14:15:11.275613Z","shell.execute_reply.started":"2025-12-22T14:15:11.271489Z","shell.execute_reply":"2025-12-22T14:15:11.274805Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n

\n ECG image processor: preprocessing + extraction of 12-lead signals\n

","metadata":{}},{"cell_type":"code","source":"class ECGImageProcessor:\n \n def __init__(self):\n self.lead_positions = {\n 'I': (0.1, 0.15), 'II': (0.1, 0.3), 'III': (0.1, 0.45),\n 'aVR': (0.1, 0.6), 'aVL': (0.1, 0.75), 'aVF': (0.1, 0.9),\n 'V1': (0.55, 0.15), 'V2': (0.55, 0.3), 'V3': (0.55, 0.45),\n 'V4': (0.55, 0.6), 'V5': (0.55, 0.75), 'V6': (0.55, 0.9)\n }\n \n def preprocess_image(self, image):\n \"\"\"Basic ECG image preprocessing.\"\"\"\n # Convert to grayscale\n if len(image.shape) == 3:\n gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)\n else:\n gray = image\n \n # Enhance contrast\n clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))\n enhanced = clahe.apply(gray)\n \n # Remove grid lines\n kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (2,2))\n opened = cv2.morphologyEx(enhanced, cv2.MORPH_OPEN, kernel)\n \n # Binarize\n _, binary = cv2.threshold(opened, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)\n \n return binary\n \n def extract_lead_signal(self, image, lead_name, fs=500):\n \"\"\"Extracting a specific lead signal.\"\"\"\n try:\n # Get lead position\n h, w = image.shape[:2]\n x_ratio, y_ratio = self.lead_positions[lead_name]\n lead_x = int(w * x_ratio)\n lead_y = int(h * y_ratio)\n \n # Define ROI around lead\n roi_width = int(w * 0.4)\n roi_height = int(h * 0.08)\n roi_x = max(0, lead_x - roi_width//2)\n roi_y = max(0, lead_y - roi_height//2)\n \n roi = image[roi_y:roi_y+roi_height, roi_x:roi_x+roi_width]\n \n if roi.size == 0:\n return np.zeros(config.target_length)\n \n # Find signal line (dark pixels)\n signal_y = []\n for col in range(roi.shape[1]):\n column = roi[:, col]\n dark_pixels = np.where(column < 128)[0]\n if len(dark_pixels) > 0:\n signal_y.append(np.mean(dark_pixels))\n else:\n signal_y.append(roi.shape[0] / 2)\n \n if not signal_y:\n return np.zeros(config.target_length)\n \n # Convert to signal\n ecg_signal = np.array(signal_y)\n \n # Invert and normalize\n ecg_signal = roi_height - ecg_signal # Invert y-axis\n ecg_signal = (ecg_signal - ecg_signal.mean()) / (ecg_signal.std() + 1e-8)\n \n # Resample to target length\n if len(ecg_signal) > 0:\n ecg_signal = scipy_signal.resample(ecg_signal, config.target_length)\n \n return ecg_signal.astype(np.float32)\n \n except Exception as e:\n print(f\"Error extracting {lead_name}: {e}\")\n return np.zeros(config.target_length)","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:15:11.276450Z","iopub.execute_input":"2025-12-22T14:15:11.276669Z","iopub.status.idle":"2025-12-22T14:15:11.419103Z","shell.execute_reply.started":"2025-12-22T14:15:11.276654Z","shell.execute_reply":"2025-12-22T14:15:11.418507Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n

\n Electrocardiography Dataset\n

","metadata":{}},{"cell_type":"code","source":"class ECGDataset(Dataset):\n \n def __init__(self, df, image_dir, transform=None, is_train=True):\n self.df = df\n self.image_dir = image_dir\n self.transform = transform\n self.is_train = is_train\n self.processor = ECGImageProcessor()\n self.leads = ['I', 'II', 'III', 'aVR', 'aVL', 'aVF', 'V1', 'V2', 'V3', 'V4', 'V5', 'V6']\n \n def __len__(self):\n return len(self.df)\n \n def __getitem__(self, idx):\n row = self.df.iloc[idx]\n base_id = row['id']\n \n # Load image\n if self.is_train:\n img_path = os.path.join(self.image_dir, str(base_id), f\"{base_id}-0001.png\")\n else:\n img_path = os.path.join(self.image_dir, f\"{base_id}.png\")\n \n image = cv2.imread(img_path)\n if image is None:\n # Try alternative images\n if self.is_train:\n for seg in ['0003', '0004', '0005']:\n img_path = os.path.join(self.image_dir, str(base_id), f\"{base_id}-{seg}.png\")\n image = cv2.imread(img_path)\n if image is not None:\n break\n \n if image is None:\n # Create dummy image as last resort\n image = np.ones((1240, 1024, 3), dtype=np.uint8) * 255 # img_size\n print(f\"Could not load image for {base_id}\")\n \n image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)\n \n if self.is_train:\n # Load ground truth from CSV\n csv_path = os.path.join(self.image_dir, str(base_id), f\"{base_id}.csv\")\n try:\n signals_df = pd.read_csv(csv_path)\n # Use lead II as target for training\n target_signal = signals_df['II'].values.astype(np.float32)\n \n # Resize to target length\n if len(target_signal) > config.target_length:\n target_signal = target_signal[:config.target_length]\n else:\n target_signal = np.pad(target_signal, (0, config.target_length - len(target_signal)), \n mode='constant')\n \n # Normalize\n if target_signal.std() > 0:\n target_signal = (target_signal - target_signal.mean()) / target_signal.std()\n \n except Exception as e:\n print(f\"Error loading CSV for {base_id}: {e}\")\n # Create synthetic ECG as fallback\n t = np.linspace(0, 10, config.target_length)\n target_signal = (np.sin(2 * np.pi * 1 * t) + \n 0.5 * np.sin(2 * np.pi * 2 * t) +\n 0.2 * np.sin(2 * np.pi * 0.5 * t))\n target_signal = target_signal.astype(np.float32)\n \n # Process image to extract features\n processed_img = self.processor.preprocess_image(image_rgb)\n \n # Extract lead II signal from image (this will be our input feature)\n extracted_signal = self.processor.extract_lead_signal(processed_img, 'II')\n \n # Prepare image for CNN\n if self.transform:\n image_tensor = self.transform(image=image_rgb)['image']\n else:\n # Default transform\n image_tensor = torch.from_numpy(\n cv2.resize(image_rgb, config.img_size).transpose(2, 0, 1)\n ).float() / 255.0\n \n return image_tensor, torch.FloatTensor(extracted_signal), torch.FloatTensor(target_signal), base_id\n \n else:\n # For test - just return image\n if self.transform:\n image_tensor = self.transform(image=image_rgb)['image']\n else:\n image_tensor = torch.from_numpy(\n cv2.resize(image_rgb, config.img_size).transpose(2, 0, 1)\n ).float() / 255.0\n \n return image_tensor, base_id","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:15:11.420846Z","iopub.execute_input":"2025-12-22T14:15:11.421104Z","iopub.status.idle":"2025-12-22T14:15:11.433843Z","shell.execute_reply.started":"2025-12-22T14:15:11.421087Z","shell.execute_reply":"2025-12-22T14:15:11.433247Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n

\n Model architecture: Image to Signal Regression\n

","metadata":{}},{"cell_type":"code","source":"class ECGNet(nn.Module):\n \n def __init__(self):\n super().__init__()\n \n # CNN for image features\n self.cnn = nn.Sequential(\n # Block 1\n nn.Conv2d(3, 64, 3, padding=1),\n nn.BatchNorm2d(64),\n nn.ReLU(),\n nn.Conv2d(64, 64, 3, padding=1),\n nn.BatchNorm2d(64),\n nn.ReLU(),\n nn.MaxPool2d(2),\n \n # Block 2 \n nn.Conv2d(64, 128, 3, padding=1),\n nn.BatchNorm2d(128),\n nn.ReLU(),\n nn.Conv2d(128, 128, 3, padding=1),\n nn.BatchNorm2d(128),\n nn.ReLU(),\n nn.MaxPool2d(2),\n \n # Block 3\n nn.Conv2d(128, 256, 3, padding=1),\n nn.BatchNorm2d(256),\n nn.ReLU(),\n nn.Conv2d(256, 256, 3, padding=1),\n nn.BatchNorm2d(256),\n nn.ReLU(),\n nn.AdaptiveAvgPool2d((4, 4))\n )\n \n # Regression_head\n self.regressor = nn.Sequential(\n nn.Linear(256 * 4 * 4, 1024),\n nn.ReLU(),\n nn.Dropout(0.5),\n nn.Linear(1024, 512),\n nn.ReLU(), \n nn.Dropout(0.3),\n nn.Linear(512, config.target_length)\n )\n \n def forward(self, x):\n # CNN features\n features = self.cnn(x)\n features = features.view(features.size(0), -1)\n \n # Regression\n output = self.regressor(features)\n return output","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:15:11.434533Z","iopub.execute_input":"2025-12-22T14:15:11.434745Z","iopub.status.idle":"2025-12-22T14:15:11.446152Z","shell.execute_reply.started":"2025-12-22T14:15:11.434721Z","shell.execute_reply":"2025-12-22T14:15:11.445455Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n

\n ECG Loss Function\n

","metadata":{}},{"cell_type":"code","source":"class ECGLoss(nn.Module):\n \n def __init__(self, eps=1e-8):\n super().__init__()\n self.eps = eps\n\n def forward(self, pred, target):\n snr = self.snr_loss(pred, target)\n smooth = self.smoothness_loss(pred)\n return snr + 0.1 * smooth # smoothness as a regularizer\n\n def snr_loss(self, pred, target):\n noise = target - pred\n \n signal_power = torch.sum(target ** 2, dim=1)\n noise_power = torch.sum(noise ** 2, dim=1)\n \n snr = signal_power / (noise_power + self.eps)\n \n return -torch.mean(torch.log(snr + self.eps))\n\n def smoothness_loss(self, signal):\n if signal.size(1) < 3:\n return torch.tensor(0.0, device=signal.device)\n \n diff1 = signal[:, 1:] - signal[:, :-1]\n diff2 = diff1[:, 1:] - diff1[:, :-1]\n \n return torch.mean(diff2 ** 2)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:15:11.446928Z","iopub.execute_input":"2025-12-22T14:15:11.447163Z","iopub.status.idle":"2025-12-22T14:15:11.458876Z","shell.execute_reply.started":"2025-12-22T14:15:11.447138Z","shell.execute_reply":"2025-12-22T14:15:11.458159Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n

\n Model training\n

","metadata":{}},{"cell_type":"code","source":"train_df = pd.read_csv('/kaggle/input/physionet-ecg-image-digitization/train.csv')\ntest_df = pd.read_csv('/kaggle/input/physionet-ecg-image-digitization/test.csv')\n\ntrain_df = train_df.head(500) # higher -> better results\n \n# Transforms\ntrain_transform = A.Compose([\n A.Resize(*config.img_size),\n A.HorizontalFlip(p=0.3),\n A.ShiftScaleRotate(shift_limit=0.05, scale_limit=0.1, rotate_limit=5, p=0.3),\n A.GridDistortion(p=0.1), # imitation of paper distortion\n A.GaussNoise(p=0.2),\n A.RandomBrightnessContrast(p=0.3),\n A.MotionBlur(p=0.1), # simulate blurring when photographing\n A.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),\n ToTensorV2()\n])\n \n# Dataset\ntrain_dataset = ECGDataset(\n train_df, \n '/kaggle/input/physionet-ecg-image-digitization/train',\n transform=train_transform,\n is_train=True\n)\n \ntrain_loader = DataLoader(\n train_dataset, \n batch_size=config.batch_size, \n shuffle=True, \n num_workers=config.num_workers\n)\n \n# Model\nmodel = ECGNet()\n\n# Two GPU's\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\nif torch.cuda.device_count() > 1:\n print(f\"Using {torch.cuda.device_count()} GPU's\")\n model = nn.DataParallel(model)\n\nmodel = model.to(device)\n \n# Loss and optimizer\ncriterion = ECGLoss()\noptimizer = optim.AdamW(model.parameters(), lr=config.lr, weight_decay=1e-4)\nscheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, patience=3, factor=0.5)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:15:11.459578Z","iopub.execute_input":"2025-12-22T14:15:11.459779Z","iopub.status.idle":"2025-12-22T14:17:13.355057Z","shell.execute_reply.started":"2025-12-22T14:15:11.459765Z","shell.execute_reply":"2025-12-22T14:17:13.353983Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"best_loss = float('inf')\nfor epoch in range(config.epochs):\n model.train()\n running_loss = 0.0\n \n for batch_idx, (images, extracted, targets, base_ids) in enumerate(train_loader):\n images = images.to(device)\n targets = targets.to(device)\n \n optimizer.zero_grad()\n outputs = model(images)\n loss = criterion(outputs, targets)\n loss.backward()\n \n torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)\n optimizer.step()\n \n running_loss += loss.item()\n \n epoch_loss = running_loss / len(train_loader)\n scheduler.step(epoch_loss)\n \n print(f'Epoch {epoch+1}/{config.epochs}, Loss: {epoch_loss:.4f}')\n \n if epoch_loss < best_loss:\n best_loss = epoch_loss\n torch.save(model.state_dict(), 'best_ecg_model.pth')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n

\n Predict, create submission and check results\n

","metadata":{}},{"cell_type":"code","source":"from scipy.signal import butter, filtfilt\n\ndef smooth_ecg(x, fs=500, lowcut=0.5, highcut=40):\n nyq = 0.5 * fs\n b, a = butter(2, [lowcut / nyq, highcut / nyq], btype='band')\n return filtfilt(b, a, x)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model.load_state_dict(torch.load('best_ecg_model.pth', map_location=device)) # Load best weights\ntest_df = pd.read_csv('/kaggle/input/physionet-ecg-image-digitization/test.csv')\n \n# Test transform\ntest_transform = A.Compose([\n A.Resize(*config.img_size),\n A.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),\n ToTensorV2()\n])\n \ntest_dataset = ECGDataset(\n test_df, \n '/kaggle/input/physionet-ecg-image-digitization/test',\n transform=test_transform,\n is_train=False\n)\n \ntest_loader = DataLoader(test_dataset, batch_size=1, shuffle=False)\n \nsubmission_data = []\nprocessor = ECGImageProcessor()\n \nmodel.eval()\n \nfor batch_idx, (images, base_ids) in enumerate(test_loader):\n if batch_idx >= len(test_df): # Safety check\n break\n \n test_row = test_df.iloc[batch_idx]\n base_id = test_row['id']\n lead = test_row['lead']\n num_rows = test_row['number_of_rows']\n \n images = images.to(device)\n \n with torch.no_grad():\n prediction = model(images).cpu().numpy().flatten()\n \n # Adjust length to required number of rows\n if len(prediction) > num_rows:\n prediction = prediction[:num_rows]\n elif len(prediction) < num_rows:\n prediction = np.pad(prediction, (0, num_rows - len(prediction)), mode='edge')\n \n prediction = smooth_ecg(prediction, fs=test_row['fs']) # New feature\n \n for row_id in range(num_rows):\n composite_id = f\"{base_id}_{row_id}_{lead}\"\n submission_data.append({\n 'id': composite_id,\n 'value': float(prediction[row_id])\n })\n\nsubmission = pd.DataFrame(submission_data)\nsubmission.to_csv('submission1.csv', index=False)\nsubmission.head(30)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:17:13.356901Z","iopub.execute_input":"2025-12-22T14:17:13.357579Z","iopub.status.idle":"2025-12-22T14:17:15.466561Z","shell.execute_reply.started":"2025-12-22T14:17:13.357552Z","shell.execute_reply":"2025-12-22T14:17:15.465759Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"## second model","metadata":{}},{"cell_type":"code","source":"!pip uninstall -y tensorflow\n!uv pip install --no-deps --system --no-index --find-links='/kaggle/input/hengck23-submit-physionet/hengck23-submit-physionet/setup' 'connected-components-3d'","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:17:15.467476Z","iopub.execute_input":"2025-12-22T14:17:15.467842Z","iopub.status.idle":"2025-12-22T14:17:37.527079Z","shell.execute_reply.started":"2025-12-22T14:17:15.467817Z","shell.execute_reply":"2025-12-22T14:17:37.526414Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import kagglehub\ndeterministic = kagglehub.package_import('wasupandceacar/deterministic').deterministic\ndeterministic.init_all(35, disable_list=['cuda_block'])\n\nimport sys\nsys.path.append('/kaggle/input/hengck23-submit-physionet/hengck23-submit-physionet')\n\nimport os\nimport gc\nimport cv2\nimport torch\nimport traceback\nimport numpy as np\nimport pandas as pd\n\nfrom tqdm import tqdm\nfrom pathlib import Path\nfrom tqdm.auto import tqdm\nfrom shutil import copyfile\nfrom scipy.signal import resample\n\nfrom stage0_model import Net as Stage0Net\nfrom stage0_common import *\n\nif_submit = os.getenv('KAGGLE_IS_COMPETITION_RERUN')\n\nif if_submit:\n test_meta = Path(\"/kaggle/input/physionet-ecg-image-digitization/test.csv\")\n test_dir = Path(\"/kaggle/input/physionet-ecg-image-digitization/test\")\nelse:\n test_meta = Path(\"/kaggle/input/physio-test-fake-dataset/test_fake/test.csv\")\n test_dir = Path(\"/kaggle/input/physio-test-fake-dataset/test_fake\")\n\nvalid_df = pd.read_csv(test_meta)\nvalid_df['id'] = valid_df['id'].astype(str) \nvalid_id = valid_df['id'].unique().tolist()\n\n\nglobal_dict = {\n \"stage0_dir\": \"/kaggle/working/stage0\",\n \"stage1_dir\": \"/kaggle/working/stage1\",\n \"stage2_dir\": \"/kaggle/working/stage2\",\n}\n\ndef change_color(image_rgb):\n gray = cv2.cvtColor(image_rgb, cv2.COLOR_RGB2GRAY)\n denoised = cv2.fastNlMeansDenoising(gray, h=10)\n clahe = cv2.createCLAHE(clipLimit=4.0, tileGridSize=(8,8))\n contrast_enhanced = clahe.apply(denoised)\n return cv2.cvtColor(contrast_enhanced, cv2.COLOR_GRAY2RGB)\n\nstage0_dir = Path(global_dict[\"stage0_dir\"])\nstage0_dir.mkdir(exist_ok=True)\n\nstage0_net = Stage0Net(pretrained=False)\nstage0_net = load_net(stage0_net, '/kaggle/input/hengck23-submit-physionet/hengck23-submit-physionet/weight/stage0-last.checkpoint.pth')\nstage0_net.to(\"cuda:0\")\nstage0_net.eval()\n\nfor n, sample_id in enumerate(tqdm(valid_id)):\n path = test_dir / f'{sample_id}.png'\n output_path = stage0_dir / f'{sample_id}.png'\n \n image_original = cv2.imread(str(path), cv2.IMREAD_COLOR)\n image_original = cv2.cvtColor(image_original, cv2.COLOR_BGR2RGB)\n image_for_model = change_color(image_original)\n \n batch = image_to_batch(image_for_model)\n\n try:\n with torch.no_grad(), torch.amp.autocast('cuda', dtype=torch.float32):\n output = stage0_net(batch)\n \n rotated, keypoint = output_to_predict(image_original, batch, output)\n normalised, _, _ = normalise_by_homography(rotated, keypoint)\n \n cv2.imwrite(str(output_path), cv2.cvtColor(normalised, cv2.COLOR_RGB2BGR))\n except Exception as e:\n traceback.print_exc()\n copyfile(path, output_path)\n\n\nfrom stage1_model import Net as Stage1Net\nfrom stage1_common import *\n\nstage0_dir = Path(global_dict[\"stage0_dir\"])\nstage1_dir = Path(global_dict[\"stage1_dir\"])\nstage1_dir.mkdir(exist_ok=True)\n\nstage1_net = Stage1Net(pretrained=False)\nstage1_net = load_net(stage1_net, '/kaggle/input/hengck23-submit-physionet/hengck23-submit-physionet/weight/stage1-last.checkpoint.pth')\nstage1_net.to(\"cuda:0\")\n\nfor n, sample_id in enumerate(tqdm(valid_id)):\n path = stage0_dir / f'{sample_id}.png'\n output_path = stage1_dir / f'{sample_id}.png'\n image = cv2.imread(path, cv2.IMREAD_COLOR_RGB)\n batch = {'image': torch.from_numpy(np.ascontiguousarray(image.transpose(2, 0, 1))).unsqueeze(0)}\n\n try:\n with torch.no_grad(), torch.amp.autocast('cuda', dtype=torch.float32):\n output = stage1_net(batch)\n gridpoint_xy, _ = output_to_predict(image, batch, output)\n rectified = rectify_image(image, gridpoint_xy)\n cv2.imwrite(output_path, cv2.cvtColor(rectified, cv2.COLOR_RGB2BGR))\n except:\n traceback.print_exc()\n copyfile(path, output_path)\n\nimport torchvision.transforms as T\nfrom stage2_model import *\nfrom stage2_common import *\nfrom scipy.signal import savgol_filter, medfilt\n\n\nclass Net3(nn.Module):\n \n def __init__(self, pretrained=True):\n super(Net3, self).__init__()\n encoder_dim = [64, 128, 256, 512]\n decoder_dim = [128, 64, 32, 16]\n\n self.encoder = timm.create_model(\n model_name='resnet34.a3_in1k', pretrained=pretrained, in_chans=3, num_classes=0, global_pool=''\n )\n\n self.decoder = MyCoordUnetDecoder(\n in_channel=encoder_dim[-1],\n skip_channel=encoder_dim[:-1][::-1] + [0],\n out_channel=decoder_dim,\n scale=[2, 2, 2, 2]\n )\n self.pixel = nn.Conv2d(decoder_dim[-1], 4, 1)\n\n def forward(self, image):\n encode = encode_with_resnet(self.encoder, image)\n last, _ = self.decoder(feature=encode[-1], skip=encode[:-1][::-1] + [None])\n pixel = self.pixel(last)\n return pixel\n\nstage1_dir = Path(global_dict[\"stage1_dir\"])\nstage2_dir = Path(global_dict[\"stage2_dir\"])\nstage2_dir.mkdir(exist_ok=True)\n\nstage2_net = Net3(pretrained=False).to(\"cuda:0\")\nmodel_path = \"/kaggle/input/physio-seg-public/pytorch/net3_009_4200/1/iter_0004200.pt\"\nstage2_net.load_state_dict(torch.load(model_path))\nstage2_net.eval()\n\nx0, x1 = 0, 2176\ny0, y1 = 0, 1696\nzero_mv = [703.5, 987.5, 1271.5, 1531.5]\nmv_to_pixel = 78.5\nt0, t1 = 235, 4161\n\nresize = T.Resize((1696, 4352), interpolation=T.InterpolationMode.BILINEAR)\n\nfor n, sample_id in enumerate(tqdm(valid_id)):\n path = stage1_dir / f'{sample_id}.png'\n output_path = stage2_dir / f'{sample_id}.npy'\n image = cv2.imread(path, cv2.IMREAD_COLOR_RGB)\n \n length = valid_df[(valid_df['id']==sample_id) & (valid_df['lead']=='II')].iloc[0].number_of_rows\n \n image = image[y0:y1, x0:x1] / 255\n batch = resize(torch.from_numpy(np.ascontiguousarray(image.transpose(2, 0, 1))).unsqueeze(0)).float().to(\"cuda:0\")\n \n try:\n with torch.no_grad(), torch.amp.autocast('cuda', dtype=torch.float32):\n output = stage2_net(batch)\n \n pixel = torch.sigmoid(output).float().data.cpu().numpy()[0]\n series_in_pixel = pixel_to_series(pixel[..., t0:t1], zero_mv, length)\n series = (np.array(zero_mv).reshape(4, 1) - series_in_pixel) / mv_to_pixel\n \n \n for i in range(series.shape[0]):\n series[i] = savgol_filter(series[i], window_length=7, polyorder=2)\n\n np.save(output_path, series)\n except:\n traceback.print_exc()\n series = np.zeros((4, length)) \n np.save(output_path, series)\n\ndef series_dict(series):\n d = {}\n for l in range(3):\n lead_names = [\n ['I', 'aVR', 'V1', 'V4'],\n ['II', 'aVL', 'V2', 'V5'],\n ['III', 'aVF', 'V3', 'V6'],\n ][l]\n split = np.array_split(series[l], 4)\n for (k, s) in zip(lead_names, split):\n d[k] = s\n \n d['II'] = series[3]\n \n return d\n\n\nstage2_dir = Path(global_dict[\"stage2_dir\"])\n\nres = []\ngb = valid_df.groupby('id')\n\nfor i, (sample_id, df) in enumerate(tqdm(gb)):\n series = np.load(stage2_dir / f'{sample_id}.npy')\n d_series = series_dict(series)\n\n for _, d in df.iterrows():\n s = d_series.get(d.lead, np.zeros(d.number_of_rows))\n \n if len(s) != d.number_of_rows:\n x_old = np.linspace(0, 1, len(s))\n x_new = np.linspace(0, 1, d.number_of_rows)\n s = np.interp(x_new, x_old, s)\n \n row_id = [f'{sample_id}_{x}_{d.lead}' for x in range(d.number_of_rows)]\n res.append(pd.DataFrame({'id': row_id, 'value': s}))\n\n if i % 100 == 0:\n gc.collect()\n\nsubmission = pd.concat(res, axis=0, ignore_index=True)\nsubmission.to_csv('submission2.csv', index=False)\nsubmission.head(30)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:17:37.529401Z","iopub.execute_input":"2025-12-22T14:17:37.529624Z","iopub.status.idle":"2025-12-22T14:18:42.935850Z","shell.execute_reply.started":"2025-12-22T14:17:37.529606Z","shell.execute_reply":"2025-12-22T14:18:42.935042Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"sub1 = pd.read_csv('submission1.csv')\nsub2 = pd.read_csv('submission2.csv')\n\neps = 1e-69\n\nweight_sub1 = eps\nweight_sub2 = 1 - eps\n\nsub2['value'] = weight_sub1 * sub1['value'] + weight_sub2 * sub2['value']\nsub2.to_csv('submission.csv', index=False)\nsub2.head(30)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-22T14:19:10.115543Z","iopub.execute_input":"2025-12-22T14:19:10.115825Z","iopub.status.idle":"2025-12-22T14:19:10.588009Z","shell.execute_reply.started":"2025-12-22T14:19:10.115807Z","shell.execute_reply":"2025-12-22T14:19:10.587211Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}