![\"render\"](\"https://storage.googleapis.com/kaggle-forum-message-attachments/2556712/20050/render.jpg\")
I find it helpful to gain different perspectives of the data we're analyzing. Napari can be helpful for visualization. (Thanks, Bull!) In addition, here's a Python script that converts a set of labels into a point cloud (point_cloud.ply) and 3D model (mesh.obj), so we can see the whole data set at once from different angles.
\nThis script takes about three minutes and 15 GB of RAM to process kidney_1_dense (Ubuntu 22.04, AMD Ryzen 5600 mobile, no dedicated GPU)
os\n numpy np\n matplotlib.pyplot plt\n skimage io\n\n\ndirectory = \n\n\ntiff_files = ([f f os.listdir(directory) f.endswith()])\n\n\n()\nimages = [io.imread(os.path.join(directory, file)).astype(np.uint8) file tiff_files]\n\n\nfirst_image_shape = images[].shape\nfirst_image_dtype = images[].dtype\n\nfirst_image_shape, first_image_dtype, tiff_files\n\n\n()\nbinary_images = [np.where(image > , , ).astype(np.uint8) image images] \nvoxel_grid = np.stack(binary_images, axis=) \n\n\n images\n binary_images\n\n\n\n\n\n\n()\nx, y, z = voxel_grid.nonzero()\n\n\n\n\n\n\n\n\n\n\n\n\n skimage.measure marching_cubes\n open3d o3d\n\n\n()\nverts, faces, _, _ = marching_cubes(voxel_grid, level=)\n\n\n voxel_grid\n\n\npoint_cloud = o3d.geometry.PointCloud()\npoint_cloud.points = o3d.utility.Vector3dVector(np.column_stack((x, y, z)))\n\n\n()\nmesh = o3d.geometry.TriangleMesh()\nmesh.vertices = o3d.utility.Vector3dVector(verts)\nmesh.triangles = o3d.utility.Vector3iVector(faces)\n\n\npoint_cloud_file = \nmesh_file = \n\n\n()\no3d.io.write_point_cloud(point_cloud_file, point_cloud)\n()\no3d.io.write_triangle_mesh(mesh_file, mesh)\n\n()\npoint_cloud_file, mesh_file\nHere is a render of the kidney_3_dense label folder.
\n
I find it helpful to gain different perspectives of the data we're analyzing. Napari can be helpful for visualization. (Thanks, Bull!) In addition, here's a Python script that converts a set of labels into a point cloud (point_cloud.ply) and 3D model (mesh.obj), so we can see the whole data set at once from different angles.
\nThis script takes about three minutes and 15 GB of RAM to process kidney_1_dense (Ubuntu 22.04, AMD Ryzen 5600 mobile, no dedicated GPU)
os\n numpy np\n matplotlib.pyplot plt\n skimage io\n\n\ndirectory = \n\n\ntiff_files = ([f f os.listdir(directory) f.endswith()])\n\n\n()\nimages = [io.imread(os.path.join(directory, file)).astype(np.uint8) file tiff_files]\n\n\nfirst_image_shape = images[].shape\nfirst_image_dtype = images[].dtype\n\nfirst_image_shape, first_image_dtype, tiff_files\n\n\n()\nbinary_images = [np.where(image > , , ).astype(np.uint8) image images] \nvoxel_grid = np.stack(binary_images, axis=) \n\n\n images\n binary_images\n\n\n\n\n\n\n()\nx, y, z = voxel_grid.nonzero()\n\n\n\n\n\n\n\n\n\n\n\n\n skimage.measure marching_cubes\n open3d o3d\n\n\n()\nverts, faces, _, _ = marching_cubes(voxel_grid, level=)\n\n\n voxel_grid\n\n\npoint_cloud = o3d.geometry.PointCloud()\npoint_cloud.points = o3d.utility.Vector3dVector(np.column_stack((x, y, z)))\n\n\n()\nmesh = o3d.geometry.TriangleMesh()\nmesh.vertices = o3d.utility.Vector3dVector(verts)\nmesh.triangles = o3d.utility.Vector3iVector(faces)\n\n\npoint_cloud_file = \nmesh_file = \n\n\n()\no3d.io.write_point_cloud(point_cloud_file, point_cloud)\n()\no3d.io.write_triangle_mesh(mesh_file, mesh)\n\n()\npoint_cloud_file, mesh_file\nHere is a render of the kidney_3_dense label folder.
\n
good work.
\ni am interested to know how to setup the render.
\nany scripts for that?
I wish there were. All the rendering work was manual
", "rawMarkdown": "I wish there were. All the rendering work was manual", "votes": null }, { "id": "2557852", "postDate": "12/11/2023 17:44:40", "content": "Cinematic rendering from simens healthineers …
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2Fca2866ee76b7ebd8ff69060ec8e7c6dc%2FSelection_999(4366).png?generation=1702316644464135&alt=media\")
Here's a template Blender project you can use. It comes with a matte studio-like background with two-point lighting, some simple shaders, and a little compositing pipeline. I created the file, and I'm releasing it into public domain.
\nSetting up a render can be as easy as opening the file in Blender 4.0+ and swapping Suzanne the monkey with an asset of your choice.
\nEnjoy!
", "rawMarkdown": "Here's a template Blender project you can use. It comes with a matte studio-like background with two-point lighting, some simple shaders, and a little compositing pipeline. I created the file, and I'm releasing it into public domain.\n\nSetting up a render can be as easy as opening the file in Blender 4.0+ and swapping Suzanne the monkey with an asset of your choice.\n\nEnjoy!", "votes": null }, { "id": "2560897", "postDate": "12/14/2023 04:50:57", "content": "thanks a lot!
", "rawMarkdown": "thanks a lot!", "votes": null } ], "comments": [ { "id": 2556774, "author_name": "hengck23", "author_url": "", "post_date": "12/11/2023 01:24:55", "content": "good work.
\ni am interested to know how to setup the render.
\nany scripts for that?
I wish there were. All the rendering work was manual
", "votes": null, "replies": [ { "id": 2557852, "author_name": "hengck23", "author_url": "", "post_date": "12/11/2023 17:44:40", "content": "Cinematic rendering from simens healthineers …
\n
Here's a template Blender project you can use. It comes with a matte studio-like background with two-point lighting, some simple shaders, and a little compositing pipeline. I created the file, and I'm releasing it into public domain.
\nSetting up a render can be as easy as opening the file in Blender 4.0+ and swapping Suzanne the monkey with an asset of your choice.
\nEnjoy!
", "votes": null, "replies": [ { "id": 2560897, "author_name": "hengck23", "author_url": "", "post_date": "12/14/2023 04:50:57", "content": "thanks a lot!
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2556712": "I find it helpful to gain different perspectives of the data we're analyzing. [Napari](https://napari.org/) can be helpful for visualization. ([Thanks, Bull!](https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/461920)) In addition, here's a Python script that converts a set of labels into a point cloud (point_cloud.ply) and 3D model (mesh.obj), so we can see the whole data set at once from different angles.\n\nThis script takes about three minutes and ~~50+~~ 15 GB of RAM to process `kidney_1_dense` (Ubuntu 22.04, AMD Ryzen 5600 mobile, no dedicated GPU)\n\n```py\nimport os\nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom skimage import io\n\n# Directory containing the TIFF files\ndirectory = 'train/kidney_1_dense/labels'\n\n# List all TIFF files in the directory\ntiff_files = sorted([f for f in os.listdir(directory) if f.endswith('.tif')])\n\n# Load the images and convert them to numpy arrays\nprint(\"Loading images...\")\nimages = [io.imread(os.path.join(directory, file)).astype(np.uint8) for file in tiff_files]\n\n# Inspecting the first image dimensions and data type\nfirst_image_shape = images[0].shape\nfirst_image_dtype = images[0].dtype\n\nfirst_image_shape, first_image_dtype, tiff_files\n\n# Convert each image to a binary array and stack them to form a 3D array\nprint(\"Converting to binary...\")\nbinary_images = [np.where(image > 0, 1, 0).astype(np.uint8) for image in images] # Convert to binary\nvoxel_grid = np.stack(binary_images, axis=0) # Stack along a new axis\n\n# Delete images to save space\ndel images\ndel binary_images\n\n# # Visualizing the 3D structure\n# fig = plt.figure(figsize=(10, 10))\n# ax = fig.add_subplot(111, projection='3d')\n\n# # Plotting each voxel\nprint(\"Plotting...\")\nx, y, z = voxel_grid.nonzero()\n# ax.scatter(z, x, y, zdir='z', c='blue', marker='.', alpha=0.5)\n\n# # Setting labels and title\n# ax.set_xlabel('Z Axis')\n# ax.set_ylabel('X Axis')\n# ax.set_zlabel('Y Axis')\n# ax.set_title('3D Voxel Visualization')\n\n# # Show plot\n# plt.show()\n\n\nfrom skimage.measure import marching_cubes\nimport open3d as o3d\n\n# Generate a mesh using marching cubes algorithm\nprint(\"Applying marching cubes on the voxel grid...\")\nverts, faces, _, _ = marching_cubes(voxel_grid, level=0)\n\n# Delete the voxel grid to save space\ndel voxel_grid\n\n# Creating Open3D point cloud from non-zero voxel points\npoint_cloud = o3d.geometry.PointCloud()\npoint_cloud.points = o3d.utility.Vector3dVector(np.column_stack((x, y, z)))\n\n# Creating Open3D mesh from the vertices and faces\nprint(\"Creating mesh...\")\nmesh = o3d.geometry.TriangleMesh()\nmesh.vertices = o3d.utility.Vector3dVector(verts)\nmesh.triangles = o3d.utility.Vector3iVector(faces)\n\n# File paths for export\npoint_cloud_file = 'point_cloud.ply'\nmesh_file = 'mesh.obj'\n\n# Export point cloud and mesh\nprint(f\"Exporting point cloud to {point_cloud_file}...\")\no3d.io.write_point_cloud(point_cloud_file, point_cloud)\nprint(f\"Exporting mesh to {mesh_file}...\")\no3d.io.write_triangle_mesh(mesh_file, mesh)\n\nprint(\"Done!\")\npoint_cloud_file, mesh_file\n\n```\n\nHere is a render of the kidney_3_dense label folder.\n\n", "2556774": "good work.\ni am interested to know how to setup the render.\nany scripts for that?", "2557831": "I wish there were. All the rendering work was manual", "2557852": "Cinematic rendering from simens healthineers ...\n\n\n.png?generation=1702316644464135&alt=media)", "2560871": "Here's a template Blender project you can use. It comes with a matte studio-like background with two-point lighting, some simple shaders, and a little compositing pipeline. I created the file, and I'm releasing it into public domain.\n\nSetting up a render can be as easy as opening the file in Blender 4.0+ and swapping Suzanne the monkey with an asset of your choice.\n\nEnjoy!", "2560897": "thanks a lot!" }, "source": "meta" }