{"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":"none","dataSources":[{"sourceId":117682,"databundleVersionId":14443416,"sourceType":"competition"}],"dockerImageVersionId":31192,"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Simple Submission for the Vesuvius Challenge\n

\n\n

\n This notebook provides a simple and reliable baseline submission for competition. It generates masks for all test volumes, the inference time is +-2 hours.\n

\n\n

\n This code is not just a solution, but a starting point for research. \n I leave the detailed analysis and EDA to other people.\n

","metadata":{}},{"cell_type":"code","source":"import os\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport plotly.graph_objects as go\n\nfrom PIL import Image\nfrom numpy import gradient\n\n# plotly visualize\nimport plotly.offline as pyo\npyo.init_notebook_mode(connected=True)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-14T07:55:56.929318Z","iopub.execute_input":"2025-11-14T07:55:56.930288Z","iopub.status.idle":"2025-11-14T07:55:56.936234Z","shell.execute_reply.started":"2025-11-14T07:55:56.930248Z","shell.execute_reply":"2025-11-14T07:55:56.934953Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Three-dimensional representation of the internal structure of papyrus\n

","metadata":{}},{"cell_type":"code","source":"DATA_ROOT = \"/kaggle/input/vesuvius-challenge-surface-detection\"\nsample_id = \"1004283650\"\n\ndef read_tiff_as_array(tiff_path):\n \"\"\"Reads a multipage TIFF as a numpy array (D, H, W)\"\"\"\n img = Image.open(tiff_path)\n slices = []\n \n for i in range(img.n_frames):\n img.seek(i)\n frame = np.array(img)\n slices.append(frame)\n \n return np.stack(slices, axis=0)\n\nvolume = read_tiff_as_array(os.path.join(DATA_ROOT, \"train_images\", f\"{sample_id}.tif\"))\nlabel = read_tiff_as_array(os.path.join(DATA_ROOT, \"train_labels\", f\"{sample_id}.tif\"))\n\nsub = 6\nvol_simple = volume[::sub, ::sub, ::sub].astype(np.float32)\nlabel_simple = label[::sub, ::sub, ::sub]\n\nvol_norm = (vol_simple - vol_simple.min()) / (vol_simple.max() - vol_simple.min())\nthreshold = np.percentile(vol_norm, 95)\n\nZ, Y, X = vol_norm.shape\nz_grid, y_grid, x_grid = np.mgrid[0:Z, 0:Y, 0:X]\n\nfig = go.Figure(data=go.Isosurface(\n x=x_grid.flatten(),\n y=y_grid.flatten(),\n z=z_grid.flatten(),\n value=vol_norm.flatten(),\n isomin=threshold,\n isomax=vol_norm.max(),\n surface_count=1,\n colorscale='rainbow',\n opacity=0.55,\n caps=dict(x_show=False, y_show=False, z_show=False)\n))\n\nfig.update_layout(\n title=f\"Lightweight 3D Scan {sub}x: 1004283650\",\n scene=dict(\n xaxis_title='X',\n yaxis_title='Y',\n zaxis_title='Z',\n aspectmode='data'\n ),\n width=750, height=700,\n template='plotly_dark'\n)\n\nfig.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-14T07:55:57.841356Z","iopub.execute_input":"2025-11-14T07:55:57.841688Z","iopub.status.idle":"2025-11-14T07:56:00.571693Z","shell.execute_reply.started":"2025-11-14T07:55:57.841661Z","shell.execute_reply":"2025-11-14T07:56:00.569983Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Reconstruction of papyrus surface geometry\n

","metadata":{}},{"cell_type":"code","source":"z, y, x = np.where(label_simple == 1)\n\nfig = go.Figure(data=go.Scatter3d(\n x=x, y=y, z=z,\n mode='markers',\n marker=dict(\n size=2.5,\n color='red',\n opacity=0.85\n )\n))\n\nfig.update_layout(\n title=f\"Simplified Papyrus Surface: 1004283650\",\n scene=dict(\n xaxis_title='X',\n yaxis_title='Y',\n zaxis_title='Z',\n aspectmode='data'\n ),\n width=750, height=700,\n template='plotly_dark'\n)\n\nfig.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-14T07:39:33.567832Z","iopub.execute_input":"2025-11-14T07:39:33.568164Z","iopub.status.idle":"2025-11-14T07:39:33.615778Z","shell.execute_reply.started":"2025-11-14T07:39:33.568141Z","shell.execute_reply":"2025-11-14T07:39:33.614839Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Papyrus in volume: scan + surface\n

","metadata":{}},{"cell_type":"code","source":"fig = go.Figure()\n\nfig.add_trace(go.Isosurface(\n x=x_grid.flatten(),\n y=y_grid.flatten(),\n z=z_grid.flatten(),\n value=vol_norm.flatten(),\n isomin=threshold,\n isomax=vol_norm.max(),\n colorscale='hsv',\n opacity=0.3\n))\n\nfig.add_trace(go.Scatter3d(\n x=x, y=y, z=z,\n mode='markers',\n marker=dict(\n size=2,\n color=z,\n colorscale='Plasma',\n opacity=0.9\n ),\n name='Papyrus Layer'\n))\n\nfig.update_layout(\n title=f\"Scan + Surface {sub}x\",\n scene=dict(aspectmode='data'),\n width=800, height=700,\n template='plotly_dark'\n)\n\nfig.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-14T07:41:13.392478Z","iopub.execute_input":"2025-11-14T07:41:13.392834Z","iopub.status.idle":"2025-11-14T07:41:13.466360Z","shell.execute_reply.started":"2025-11-14T07:41:13.392811Z","shell.execute_reply":"2025-11-14T07:41:13.465185Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n 3D reconstruction of the median line of the scroll texture\n

","metadata":{}},{"cell_type":"code","source":"centroids = []\nfor z in range(label.shape[0]):\n mask = (label[z] == 1)\n if np.any(mask):\n y_coords, x_coords = np.where(mask)\n centroids.append([z, y_coords.mean(), x_coords.mean()])\n\ncentroids = np.array(centroids)\n\nfig = go.Figure(data=go.Scatter3d(\n x=centroids[:, 2],\n y=centroids[:, 1],\n z=centroids[:, 0],\n mode='lines+markers',\n line=dict(color='gold', width=6),\n marker=dict(size=4, color='#FFD700', symbol='circle')\n))\n\nfig.update_layout(\n title=\"3D Trajectory of Papyrus Layer (Centerline)\",\n scene=dict(\n xaxis_title='X (width)',\n yaxis_title='Y (height)',\n zaxis_title='Z (depth)',\n aspectmode='data'\n ),\n width=800, height=700,\n template='plotly_dark'\n)\n\nfig.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-14T07:43:15.968754Z","iopub.execute_input":"2025-11-14T07:43:15.969072Z","iopub.status.idle":"2025-11-14T07:43:16.209046Z","shell.execute_reply.started":"2025-11-14T07:43:15.969051Z","shell.execute_reply":"2025-11-14T07:43:16.207960Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Distribution of signal and segmentation density by scan depth\n

","metadata":{}},{"cell_type":"code","source":"signal_profile = np.mean(volume, axis=(1, 2)) # signal: average intensity per layer\nmask_profile = np.mean((label == 1).astype(float), axis=(1, 2)) # mask: the percentage of segmentation pixels per layer\n\n# normalize for comparison\nsignal_norm = (signal_profile - signal_profile.min()) / (signal_profile.max() - signal_profile.min())\nmask_norm = mask_profile / mask_profile.max()\n\nplt.figure(figsize=(12, 5))\nplt.plot(signal_norm, label='Normalized CT Signal', color='gray', linewidth=2)\nplt.fill_between(range(len(mask_norm)), mask_norm, color='red', alpha=0.4, label='Segmentation Density')\n\nplt.title(\"Signal vs Segmentation Density Along Depth (Z-axis)\")\nplt.xlabel(\"Slice Index (Z)\")\nplt.ylabel(\"Normalized Intensity / Density\")\n\nplt.legend()\nplt.grid(True, linestyle='--', alpha=0.6)\nplt.tight_layout()\n\nplt.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-14T07:43:26.249004Z","iopub.execute_input":"2025-11-14T07:43:26.249907Z","iopub.status.idle":"2025-11-14T07:43:26.784656Z","shell.execute_reply.started":"2025-11-14T07:43:26.249872Z","shell.execute_reply":"2025-11-14T07:43:26.783623Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n The gradient of the papyrus structure and mask in cross section\n

","metadata":{}},{"cell_type":"code","source":"z_mid = volume.shape[0] // 2\nimg_slice = volume[z_mid].astype(float)\nmask_slice = (label[z_mid] == 1)\n\ngrad_y, grad_x = gradient(img_slice)\nedge_strength = np.hypot(grad_x, grad_y)\n\nimg_norm = (img_slice - img_slice.min()) / (img_slice.max() - img_slice.min())\nedge_norm = (edge_strength - edge_strength.min()) / (edge_strength.max() - edge_strength.min())\n\nfig = go.Figure()\n\nfig.add_trace(go.Heatmap(\n z=edge_norm,\n colorscale='Blues',\n showscale=False,\n name='Structure'\n))\n\ny_mask, x_mask = np.where(mask_slice)\nfig.add_trace(go.Scatter(\n x=x_mask, y=y_mask,\n mode='markers',\n marker=dict(color='red', size=1.2, opacity=0.7),\n name='Papyrus'\n))\n\nfig.update_layout(\n title=f\"Structural Edges + Papyrus Mask (Z={z_mid})\",\n xaxis=dict(scaleanchor=\"y\", scaleratio=1),\n yaxis=dict(autorange=\"reversed\"),\n width=700, height=700,\n showlegend=True,\n template='plotly_dark'\n)\n\nfig.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-14T07:43:30.293210Z","iopub.execute_input":"2025-11-14T07:43:30.293617Z","iopub.status.idle":"2025-11-14T07:43:30.385968Z","shell.execute_reply.started":"2025-11-14T07:43:30.293587Z","shell.execute_reply":"2025-11-14T07:43:30.384484Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n\n
\n\n
\n\n

\n Simple submission\n

","metadata":{}},{"cell_type":"code","source":"import os\nimport zipfile\nimport numpy as np\nimport pandas as pd\nfrom PIL import Image\n\ntest_csv_path = \"/kaggle/input/vesuvius-challenge-surface-detection/test.csv\"\ntest_images_dir = \"/kaggle/input/vesuvius-challenge-surface-detection/test_images\"\n\ntest_df = pd.read_csv(test_csv_path)\ntest_ids = test_df['id'].tolist()\n\ndef read_tiff_as_array(tiff_path):\n \"\"\"Reads a multipage TIFF as a numpy array (D, H, W)\"\"\"\n img = Image.open(tiff_path)\n slices = []\n \n for i in range(img.n_frames):\n img.seek(i)\n frame = np.array(img)\n slices.append(frame)\n \n return np.stack(slices, axis=0)\n\ndef write_array_as_tiff(arr, out_path):\n \"\"\"Saves a 3D numpy array as a multi-page TIFF\"\"\"\n arr = arr.astype(np.uint8)\n images = [Image.fromarray(arr[i], mode='L') for i in range(arr.shape[0])]\n images[0].save(\n out_path,\n save_all=True,\n append_images=images[1:],\n compression=None,\n )\n\ndef create_mask(volume):\n depth, height, width = volume.shape\n \n mask = np.zeros((depth, height, width), dtype=np.uint8)\n \n depth_range = max(3, depth // 4)\n d1, d2 = depth // 2 - depth_range, depth // 2 + depth_range\n \n size_h = max(1, height // 2)\n size_w = max(1, width // 2)\n \n h1, h2 = height // 2 - size_h, height // 2 + size_h\n w1, w2 = width // 2 - size_w, width // 2 + size_w\n \n d1, d2 = max(0, d1), min(depth, d2)\n h1, h2 = max(0, h1), min(height, h2)\n w1, w2 = max(0, w1), min(width, w2)\n \n mask[d1:d2, h1:h2, w1:w2] = 1\n \n return mask\n\nfor test_id in test_ids:\n image_path = os.path.join(test_images_dir, f\"{test_id}.tif\")\n \n volume = read_tiff_as_array(image_path)\n \n mask = create_mask(volume)\n \n write_array_as_tiff(mask, f\"{test_id}.tif\")\n\nwith zipfile.ZipFile('submission.zip', 'w') as zipf:\n for test_id in test_ids:\n zipf.write(f\"{test_id}.tif\")","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true,"execution":{"iopub.status.busy":"2025-11-14T06:34:27.950292Z","iopub.execute_input":"2025-11-14T06:34:27.950686Z","iopub.status.idle":"2025-11-14T06:34:29.786823Z","shell.execute_reply.started":"2025-11-14T06:34:27.950660Z","shell.execute_reply":"2025-11-14T06:34:29.786007Z"}},"outputs":[],"execution_count":null}]}