\n\n
\n\n
\n\n
\n Create sumission models 2 Dino and ensemlbing\n
","metadata":{}},{"cell_type":"code","source":"import os\nimport cv2\nimport json\nimport math\nimport torch\nimport numpy as np\nimport pandas as pd\nimport torch.nn as nn\nimport torch.nn.functional as F\n\nfrom PIL import Image\nfrom pathlib import Path\nfrom transformers import AutoImageProcessor, AutoModel\n\nclass CONFIG:\n test_images_path = \"/kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images\"\n sample_sub_path = \"/kaggle/input/recodai-luc-scientific-image-forgery-detection/sample_submission.csv\"\n model1_path = \"/kaggle/input/modelsbest309base/best_model.pth\"\n model2_path = \"/kaggle/input/dinobestmodel/pytorch/default/1/dino197.pth\"\n dino_path = \"/kaggle/input/dinov2/pytorch/base/1\"\n device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n img_size = 512\n use_tta = True\n ensemble_strategy = \"average\"\n min_area = 150\n min_confidence = 0.33","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T17:55:50.879999Z","iopub.execute_input":"2025-12-17T17:55:50.880243Z","iopub.status.idle":"2025-12-17T17:56:17.291961Z","shell.execute_reply.started":"2025-12-17T17:55:50.880224Z","shell.execute_reply":"2025-12-17T17:56:17.291411Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class Decoder(nn.Module):\n \n def __init__(self, in_ch=768, out_ch=1):\n super().__init__()\n self.net = nn.Sequential(\n nn.Conv2d(in_ch, 256, 3, padding=1), nn.ReLU(),\n nn.Conv2d(256, 64, 3, padding=1), nn.ReLU(),\n nn.Conv2d(64, out_ch, 1)\n )\n \n def forward(self, f, size):\n return self.net(F.interpolate(f, size=size, mode=\"bilinear\", align_corners=False))","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T17:56:17.292713Z","iopub.execute_input":"2025-12-17T17:56:17.293228Z","iopub.status.idle":"2025-12-17T17:56:17.298427Z","shell.execute_reply.started":"2025-12-17T17:56:17.293198Z","shell.execute_reply":"2025-12-17T17:56:17.297738Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class DinoSegmenter(nn.Module):\n \n def __init__(self, encoder, processor):\n super().__init__()\n self.encoder = encoder\n self.processor = processor\n self.seg_head = Decoder(768, 1)\n \n def forward_features(self, x):\n imgs = (x*255).clamp(0, 255).byte().permute(0, 2, 3, 1).cpu().numpy()\n inputs = self.processor(images=list(imgs), return_tensors=\"pt\").to(x.device)\n \n with torch.no_grad():\n feats = self.encoder(**inputs).last_hidden_state\n \n B, N, C = feats.shape\n fmap = feats[:, 1:, :].permute(0, 2, 1)\n \n s = int(math.sqrt(N-1))\n fmap = fmap.reshape(B, C, s, s)\n \n return fmap\n \n def forward_seg(self, x):\n fmap = self.forward_features(x)\n return self.seg_head(fmap, (CONFIG.img_size, CONFIG.img_size))","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T17:56:17.301189Z","iopub.execute_input":"2025-12-17T17:56:17.301601Z","iopub.status.idle":"2025-12-17T17:56:17.551731Z","shell.execute_reply.started":"2025-12-17T17:56:17.301575Z","shell.execute_reply":"2025-12-17T17:56:17.550899Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class Model(nn.Module):\n \n def __init__(self):\n super().__init__()\n self.encoder = nn.Sequential(\n nn.Conv2d(3, 32, 3, padding=1), nn.ReLU(),\n nn.Conv2d(32, 32, 3, padding=1), nn.ReLU(),\n nn.MaxPool2d(2),\n nn.Conv2d(32, 64, 3, padding=1), nn.ReLU(),\n nn.Conv2d(64, 64, 3, padding=1), nn.ReLU(),\n nn.MaxPool2d(2),\n )\n \n self.decoder = nn.Sequential(\n nn.Conv2d(64, 32, 3, padding=1), nn.ReLU(),\n nn.Upsample(scale_factor=2, mode='bilinear'),\n nn.Conv2d(32, 32, 3, padding=1), nn.ReLU(),\n nn.Upsample(scale_factor=2, mode='bilinear'),\n nn.Conv2d(32, 1, 1),\n )\n \n def forward(self, x):\n x = self.encoder(x)\n x = self.decoder(x)\n x = F.interpolate(x, size=(CONFIG.img_size, CONFIG.img_size), mode='bilinear', align_corners=False)\n \n return x","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T17:56:17.552360Z","iopub.execute_input":"2025-12-17T17:56:17.552644Z","iopub.status.idle":"2025-12-17T17:56:17.562843Z","shell.execute_reply.started":"2025-12-17T17:56:17.552619Z","shell.execute_reply":"2025-12-17T17:56:17.562194Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"def load_model(model_path):\n try:\n checkpoint = torch.load(model_path, map_location=CONFIG.device)\n \n if isinstance(checkpoint, dict):\n state_dict = None\n if 'model_state_dict' in checkpoint:\n state_dict = checkpoint['model_state_dict']\n elif 'state_dict' in checkpoint:\n state_dict = checkpoint['state_dict']\n elif 'model' in checkpoint:\n model = checkpoint['model']\n if hasattr(model, 'eval'):\n model.eval()\n return model.to(CONFIG.device)\n else:\n state_dict = checkpoint\n \n try:\n processor = AutoImageProcessor.from_pretrained(CONFIG.dino_path, local_files_only=True)\n encoder = AutoModel.from_pretrained(CONFIG.dino_path, local_files_only=True).eval().to(CONFIG.device)\n model = DinoSegmenter(encoder, processor).to(CONFIG.device)\n \n if state_dict is not None:\n model.load_state_dict(state_dict, strict=False)\n \n model.eval()\n return model\n except:\n try:\n model = Model().to(CONFIG.device)\n if state_dict is not None:\n model.load_state_dict(state_dict, strict=False)\n model.eval()\n return model\n except:\n return None\n \n elif hasattr(checkpoint, 'eval'):\n checkpoint.eval()\n return checkpoint.to(CONFIG.device)\n \n return None\n except Exception as e:\n print(f\"Error {Path(model_path).name}: {e}\")\n return None\n\ndef predict_with_tta(model, image_tensor):\n predictions = []\n \n with torch.no_grad():\n if hasattr(model, 'forward_seg'):\n pred = torch.sigmoid(model.forward_seg(image_tensor))\n else:\n pred = torch.sigmoid(model(image_tensor))\n \n predictions.append(pred)\n \n with torch.no_grad():\n if hasattr(model, 'forward_seg'):\n pred = torch.sigmoid(model.forward_seg(torch.flip(image_tensor, dims=[3])))\n else:\n pred = torch.sigmoid(model(torch.flip(image_tensor, dims=[3])))\n \n predictions.append(torch.flip(pred, dims=[3]))\n \n with torch.no_grad():\n if hasattr(model, 'forward_seg'):\n pred = torch.sigmoid(model.forward_seg(torch.flip(image_tensor, dims=[2])))\n else:\n pred = torch.sigmoid(model(torch.flip(image_tensor, dims=[2])))\n \n predictions.append(torch.flip(pred, dims=[2]))\n \n if CONFIG.use_tta:\n with torch.no_grad():\n if hasattr(model, 'forward_seg'):\n pred = torch.sigmoid(model.forward_seg(torch.rot90(image_tensor, 1, [2, 3])))\n else:\n pred = torch.sigmoid(model(torch.rot90(image_tensor, 1, [2, 3])))\n \n predictions.append(torch.rot90(pred, -1, [2, 3]))\n \n return torch.stack(predictions).mean(0)[0, 0].detach().cpu().numpy()\n else:\n return predictions[0][0, 0].detach().cpu().numpy()\n\ndef postprocess(pred, original_size):\n pred = cv2.GaussianBlur(pred, (3, 3), 0)\n mean_val = np.mean(pred)\n std_val = np.std(pred)\n thr = mean_val + 0.3 * std_val\n mask = (pred > thr).astype(np.uint8)\n \n if mask.sum() > 0:\n mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, np.ones((5, 5), np.uint8))\n mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, np.ones((3, 3), np.uint8))\n \n mask = cv2.resize(mask, original_size, interpolation=cv2.INTER_NEAREST)\n return mask\n\ndef rle_encode(mask):\n pixels = mask.T.flatten()\n dots = np.where(pixels == 1)[0]\n \n if len(dots) == 0:\n return \"authentic\"\n \n run_lengths = []\n prev = -2\n \n for b in dots:\n if b > prev + 1:\n run_lengths.extend((b + 1, 0))\n run_lengths[-1] += 1\n prev = b\n \n return json.dumps([int(x) for x in run_lengths])","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T17:56:17.563540Z","iopub.execute_input":"2025-12-17T17:56:17.563827Z","iopub.status.idle":"2025-12-17T17:56:17.587104Z","shell.execute_reply.started":"2025-12-17T17:56:17.563811Z","shell.execute_reply":"2025-12-17T17:56:17.586323Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model1 = load_model(CONFIG.model1_path)\nmodel2 = load_model(CONFIG.model2_path)\n\nmodels = {}\nif model1:\n models['model1'] = model1\n print(f\"model 1 success!\")\nif model2:\n models['model2'] = model2\n print(f\"model 2 success!\")\n\n\npredictions = []\nimage_files = sorted([f for f in os.listdir(CONFIG.test_images_path) if f.lower().endswith(('.png', '.jpg', '.jpeg', '.tiff', '.bmp'))])\n\nfor image_name in image_files:\n image_path = Path(CONFIG.test_images_path) / image_name\n image = Image.open(image_path).convert(\"RGB\")\n \n image_array = np.array(image.resize((CONFIG.img_size, CONFIG.img_size)), np.float32) / 255\n image_tensor = torch.from_numpy(image_array).permute(2, 0, 1)[None].to(CONFIG.device)\n \n ensemble_preds = []\n for model in models.values():\n pred = predict_with_tta(model, image_tensor)\n ensemble_preds.append(pred)\n \n if CONFIG.ensemble_strategy == \"max\" and ensemble_preds:\n final_pred = np.max(ensemble_preds, axis=0)\n else:\n final_pred = np.mean(ensemble_preds, axis=0) if ensemble_preds else np.zeros((CONFIG.img_size, CONFIG.img_size))\n \n mask = postprocess(final_pred, image.size)\n area = int(mask.sum())\n \n if area > 0:\n mask_resized = cv2.resize(mask, (CONFIG.img_size, CONFIG.img_size), interpolation=cv2.INTER_NEAREST)\n mean_inside = float(final_pred[mask_resized == 1].mean()) if (mask_resized == 1).any() else 0.0\n else:\n mean_inside = 0.0\n \n if area < CONFIG.min_area or mean_inside < CONFIG.min_confidence:\n annotation = \"authentic\"\n else:\n annotation = rle_encode(mask)\n \n predictions.append({\n \"case_id\": Path(image_name).stem,\n \"annotation\": annotation\n })\n \npredictions_df = pd.DataFrame(predictions)\npredictions_df[\"case_id\"] = predictions_df[\"case_id\"].astype(str)\n \nsubmission = pd.read_csv(CONFIG.sample_sub_path)\nsubmission[\"case_id\"] = submission[\"case_id\"].astype(str)\nsubmission = submission[[\"case_id\"]].merge(predictions_df[[\"case_id\", \"annotation\"]], on=\"case_id\", how=\"left\")\n\nsubmission[\"annotation\"] = submission[\"annotation\"].fillna(\"authentic\")\nsubmission[[\"case_id\", \"annotation\"]].to_csv(\"submission.csv\", index=False)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T17:56:17.587935Z","iopub.execute_input":"2025-12-17T17:56:17.588173Z","iopub.status.idle":"2025-12-17T17:56:33.810028Z","shell.execute_reply.started":"2025-12-17T17:56:17.588150Z","shell.execute_reply":"2025-12-17T17:56:33.809183Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n
\n If i have mistake write pls in comments\n
","metadata":{}},{"cell_type":"code","source":"","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}