{"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":113558,"databundleVersionId":14174843,"sourceType":"competition"}],"dockerImageVersionId":31154,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n 👋 Greetings to all!\n

\n\n
\n

\n 🎯 What can be improved:\n

\n \n
\n\n

\n Good luck in improving! 🚀\n

","metadata":{}},{"cell_type":"code","source":"import os\n\nos.makedirs('/kaggle/working/ultralytics', exist_ok=True)\n\n!pip download ultralytics \"numpy<2\" --dest /kaggle/working/ultralytics","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:31:28.799409Z","iopub.execute_input":"2025-11-02T09:31:28.799864Z","iopub.status.idle":"2025-11-02T09:32:55.883217Z","shell.execute_reply.started":"2025-11-02T09:31:28.799839Z","shell.execute_reply":"2025-11-02T09:32:55.882181Z"},"_kg_hide-output":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"!pip install --no-index --find-links /kaggle/working/ultralytics ultralytics \"numpy<2\"","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:32:55.884713Z","iopub.execute_input":"2025-11-02T09:32:55.885007Z","iopub.status.idle":"2025-11-02T09:35:18.741506Z","shell.execute_reply.started":"2025-11-02T09:32:55.884981Z","shell.execute_reply":"2025-11-02T09:35:18.740802Z"},"_kg_hide-output":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Import libs\n

","metadata":{}},{"cell_type":"code","source":"import os\nimport cv2\nimport json\nimport yaml\nimport torch\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\n\nfrom PIL import Image\nfrom tqdm import tqdm\nfrom pathlib import Path\nfrom ultralytics import YOLO\nfrom sklearn.model_selection import train_test_split","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:35:18.742486Z","iopub.execute_input":"2025-11-02T09:35:18.742728Z","iopub.status.idle":"2025-11-02T09:35:21.833739Z","shell.execute_reply.started":"2025-11-02T09:35:18.742705Z","shell.execute_reply":"2025-11-02T09:35:21.833182Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Main paths\n

","metadata":{}},{"cell_type":"code","source":"train_authentic = '/kaggle/input/recodai-luc-scientific-image-forgery-detection/train_images/authentic'\ntrain_forged = '/kaggle/input/recodai-luc-scientific-image-forgery-detection/train_images/forged'\ntrain_masks = '/kaggle/input/recodai-luc-scientific-image-forgery-detection/train_masks'\n\ntest_images = '/kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images'","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:35:21.835198Z","iopub.execute_input":"2025-11-02T09:35:21.835490Z","iopub.status.idle":"2025-11-02T09:35:21.839036Z","shell.execute_reply.started":"2025-11-02T09:35:21.835475Z","shell.execute_reply":"2025-11-02T09:35:21.838285Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Create paths for learning yolo\n

","metadata":{}},{"cell_type":"code","source":"output_dir = '/kaggle/working/yolo_dataset'\nimages_dir = os.path.join(output_dir, 'images')\nlabels_dir = os.path.join(output_dir, 'labels')\n\nos.makedirs(os.path.join(images_dir, 'train'), exist_ok=True)\nos.makedirs(os.path.join(images_dir, 'val'), exist_ok=True)\nos.makedirs(os.path.join(labels_dir, 'train'), exist_ok=True)\nos.makedirs(os.path.join(labels_dir, 'val'), exist_ok=True)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:35:21.839678Z","iopub.execute_input":"2025-11-02T09:35:21.839895Z","iopub.status.idle":"2025-11-02T09:35:21.856417Z","shell.execute_reply.started":"2025-11-02T09:35:21.839879Z","shell.execute_reply":"2025-11-02T09:35:21.855923Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Read images into a single sample. Set is_forged=False for authentic images (images without a mask)\n

","metadata":{}},{"cell_type":"code","source":"samples = []\nimg_size = 640\n\n# Authentic images\nauthentic_files = [f for f in os.listdir(train_authentic) if f.endswith(('.png', '.jpg', '.jpeg'))]\nprint(f\"Found {len(authentic_files)} authentic images\")\n\nfor file in authentic_files:\n img_path = os.path.join(train_authentic, file)\n base_name = os.path.splitext(file)[0]\n \n samples.append({\n 'image_path': img_path,\n 'mask_path': None,\n 'is_forged': False,\n 'image_id': base_name\n })\n\n# Forged images\nforged_files = [f for f in os.listdir(train_forged) if f.endswith(('.png', '.jpg', '.jpeg'))]\nprint(f\"Found {len(forged_files)} forged images\")\n\nfor file in forged_files:\n img_path = os.path.join(train_forged, file)\n base_name = os.path.splitext(file)[0]\n mask_path = os.path.join(train_masks, f\"{base_name}.npy\")\n \n samples.append({\n 'image_path': img_path,\n 'mask_path': mask_path,\n 'is_forged': True,\n 'image_id': base_name\n })\n\nprint(f\"Total samples: {len(samples)}\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:35:21.857127Z","iopub.execute_input":"2025-11-02T09:35:21.857374Z","iopub.status.idle":"2025-11-02T09:35:21.931179Z","shell.execute_reply.started":"2025-11-02T09:35:21.857352Z","shell.execute_reply":"2025-11-02T09:35:21.930619Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# Split into train/val\ntrain_samples, val_samples = train_test_split(\n samples, test_size=0.1, random_state=42, \n stratify=[s['is_forged'] for s in samples]\n)\n\nprint(f\"Train samples: {len(train_samples)}\")\nprint(f\"Val samples: {len(val_samples)}\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:35:21.931822Z","iopub.execute_input":"2025-11-02T09:35:21.932040Z","iopub.status.idle":"2025-11-02T09:35:21.951330Z","shell.execute_reply.started":"2025-11-02T09:35:21.932017Z","shell.execute_reply":"2025-11-02T09:35:21.950788Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Save images and masks in output directory\n

","metadata":{}},{"cell_type":"code","source":"for sample in tqdm(train_samples):\n image_id = sample['image_id']\n \n # Resize and save image\n img = Image.open(sample['image_path']).convert('RGB')\n img_resized = img.resize((img_size, img_size), Image.Resampling.LANCZOS)\n img_save_path = os.path.join(images_dir, 'train', f\"{image_id}.jpg\")\n img_resized.save(img_save_path, 'JPEG', quality=95)\n \n # Create label file\n label_path = os.path.join(labels_dir, 'train', f\"{image_id}.txt\")\n \n if sample['is_forged'] and sample['mask_path'] and os.path.exists(sample['mask_path']):\n # Load and process mask\n mask = np.load(sample['mask_path'])\n \n # Handle multi-channel masks\n if mask.ndim == 3:\n if mask.shape[0] <= 10:\n mask = np.any(mask, axis=0)\n elif mask.shape[-1] <= 10:\n mask = np.any(mask, axis=-1)\n else:\n mask = mask[0] if mask.shape[0] <= 10 else mask[:, :, 0]\n \n mask = (mask > 0).astype(np.uint8)\n \n if mask.sum() > 0:\n # Resize mask\n mask_resized = cv2.resize(mask, (img_size, img_size), interpolation=cv2.INTER_NEAREST)\n \n # Convert to YOLO format\n contours, _ = cv2.findContours(mask_resized, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)\n \n annotations = []\n for contour in contours:\n if len(contour) > 0:\n x, y, w, h = cv2.boundingRect(contour)\n if w > 5 and h > 5:\n # Normalize coordinates\n normalized_contour = contour.astype(np.float32)\n normalized_contour[:, :, 0] /= img_size\n normalized_contour[:, :, 1] /= img_size\n \n # Create annotation string\n points = normalized_contour.reshape(-1, 2)\n points_str = ' '.join([f'{x:.6f} {y:.6f}' for x, y in points])\n annotation = f\"0 {points_str}\"\n annotations.append(annotation)\n \n # Save annotations\n with open(label_path, 'w') as f:\n for annotation in annotations:\n f.write(annotation + '\\n')\n continue\n \n # Create empty file for authentic images or images without masks\n open(label_path, 'w').close()\n\nfor sample in tqdm(val_samples):\n image_id = sample['image_id']\n \n # Resize and save image\n img = Image.open(sample['image_path']).convert('RGB')\n img_resized = img.resize((img_size, img_size), Image.Resampling.LANCZOS)\n img_save_path = os.path.join(images_dir, 'val', f\"{image_id}.jpg\")\n img_resized.save(img_save_path, 'JPEG', quality=95)\n \n # Create label file\n label_path = os.path.join(labels_dir, 'val', f\"{image_id}.txt\")\n \n if sample['is_forged'] and sample['mask_path'] and os.path.exists(sample['mask_path']):\n # Load and process mask\n mask = np.load(sample['mask_path'])\n \n # Handle multi-channel masks\n if mask.ndim == 3:\n if mask.shape[0] <= 10:\n mask = np.any(mask, axis=0)\n elif mask.shape[-1] <= 10:\n mask = np.any(mask, axis=-1)\n else:\n mask = mask[0] if mask.shape[0] <= 10 else mask[:, :, 0]\n \n mask = (mask > 0).astype(np.uint8)\n \n if mask.sum() > 0:\n # Resize mask\n mask_resized = cv2.resize(mask, (img_size, img_size), interpolation=cv2.INTER_NEAREST)\n \n # Convert to YOLO format\n contours, _ = cv2.findContours(mask_resized, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)\n \n annotations = []\n for contour in contours:\n if len(contour) > 0:\n x, y, w, h = cv2.boundingRect(contour)\n if w > 5 and h > 5:\n # Normalize coordinates\n normalized_contour = contour.astype(np.float32)\n normalized_contour[:, :, 0] /= img_size\n normalized_contour[:, :, 1] /= img_size\n \n # Create annotation string\n points = normalized_contour.reshape(-1, 2)\n points_str = ' '.join([f'{x:.6f} {y:.6f}' for x, y in points])\n annotation = f\"0 {points_str}\"\n annotations.append(annotation)\n \n # Save annotations\n with open(label_path, 'w') as f:\n for annotation in annotations:\n f.write(annotation + '\\n')\n continue\n \n # Create empty file for authentic images or images without masks\n open(label_path, 'w').close()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:35:21.952072Z","iopub.execute_input":"2025-11-02T09:35:21.952330Z","iopub.status.idle":"2025-11-02T09:41:22.784989Z","shell.execute_reply.started":"2025-11-02T09:35:21.952313Z","shell.execute_reply":"2025-11-02T09:41:22.784401Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# Create YAML config\ndata_yaml = f\"\"\"\npath: {output_dir}\n\ntrain: images/train\nval: images/val\n\nnames:\n 0: forgery\n\"\"\"\n\nwith open('data.yaml', 'w') as file:\n file.write(data_yaml)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:41:22.785802Z","iopub.execute_input":"2025-11-02T09:41:22.786076Z","iopub.status.idle":"2025-11-02T09:41:22.790693Z","shell.execute_reply.started":"2025-11-02T09:41:22.786050Z","shell.execute_reply":"2025-11-02T09:41:22.789965Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Learn Yolo11n(small) model\n

","metadata":{}},{"cell_type":"code","source":"model = YOLO('yolo11x-seg.pt')\n\nresults = model.train(\n data='data.yaml',\n epochs=50,\n imgsz=img_size, # 640\n batch=16,\n device=[0,1] if torch.cuda.is_available() else 'cpu',\n \n optimizer='Adam',\n lr0=0.001,\n lrf=0.01,\n momentum=0.9,\n weight_decay=0.05,\n \n warmup_epochs=3.0,\n warmup_momentum=0.8,\n warmup_bias_lr=0.1,\n \n hsv_h=0.015,\n hsv_s=0.7,\n hsv_v=0.4,\n degrees=45.0,\n translate=0.2,\n scale=0.5,\n shear=0.0,\n perspective=0.0001,\n flipud=0.0,\n fliplr=0.5,\n mosaic=1.0,\n mixup=0.1,\n copy_paste=0.1,\n \n box=7.5,\n cls=0.5,\n dfl=1.5,\n mask_ratio=4,\n overlap_mask=True,\n \n save=True,\n save_period=5,\n cache=False,\n workers=8,\n single_cls=False,\n rect=False,\n cos_lr=False,\n close_mosaic=10,\n \n seed=42,\n deterministic=True,\n val=True,\n plots=True,\n patience=50\n)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:41:22.792814Z","iopub.execute_input":"2025-11-02T09:41:22.793023Z","iopub.status.idle":"2025-11-02T09:45:51.575358Z","shell.execute_reply.started":"2025-11-02T09:41:22.793008Z","shell.execute_reply":"2025-11-02T09:45:51.574552Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n 🔥 Check our results\n

","metadata":{}},{"cell_type":"code","source":"def visualize_predictions(model, image_dir, counts=5, conf_threshold=0.1):\n image_paths = list(Path(image_dir).glob('*.*'))\n image_paths = [p for p in image_paths if p.suffix.lower() in {'.png', '.jpg', '.jpeg'}]\n \n if not image_paths:\n print(\"No images found in\", image_dir)\n return\n \n image_paths = image_paths[:counts]\n \n for img_path in image_paths:\n # load images\n img = cv2.imread(str(img_path))\n if img is None:\n continue\n img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n h, w = img.shape[:2]\n \n # predict\n result = model.predict(\n source=img_rgb,\n imgsz=img_size, # 512\n conf=conf_threshold,\n iou=0.45,\n verbose=False\n )[0]\n \n # create mask\n mask_vis = np.zeros((h, w), dtype=np.uint8)\n if result.masks is not None:\n for mask_tensor in result.masks.data:\n mask = mask_tensor.cpu().numpy()\n mask_resized = cv2.resize(mask, (w, h), interpolation=cv2.INTER_NEAREST)\n mask_vis = np.maximum(mask_vis, (mask_resized > 0.5).astype(np.uint8))\n \n plt.figure(figsize=(10, 5))\n plt.subplot(1, 2, 1)\n plt.imshow(img_rgb)\n plt.title(f\"{img_path.name}\")\n plt.axis('off')\n \n plt.subplot(1, 2, 2)\n plt.imshow(mask_vis, cmap='gray')\n plt.title(\"Predicted Mask\")\n plt.axis('off')\n \n plt.tight_layout()\n plt.show()\n\nvisualize_predictions(model, image_dir=train_forged, counts=3)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-02T09:45:51.576427Z","iopub.execute_input":"2025-11-02T09:45:51.576620Z","iopub.status.idle":"2025-11-02T09:45:54.305937Z","shell.execute_reply.started":"2025-11-02T09:45:51.576607Z","shell.execute_reply":"2025-11-02T09:45:54.305070Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}