![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F10530406%2F24674e6ffe8aefafcd0205ae813c5104%2FScreenshot%202024-12-03%20at%201.51.45AM.png?generation=1733208740373890&alt=media\")
I am currently creating 3d reconstructions of the cell scans. My method is as follows:
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F10530406%2Faddcd116d75262560ca388454aeef7f9%2FScreenshot%202024-12-03%20at%201.51.32AM.png?generation=1733208728485901&alt=media\")
This is a picture of a cluster of three apo-ferritin in the file TS_5_4 and you can kind of see it, but there is supposed to be a hole as shown below, so it's a little noisy. If anyone wants to try this problem out it's fun :) Also, maybe it can lead to some insights on how to solve the main goal of this competition.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F10530406%2F4361e7ef01efe8a65489971fa343215f%2F3-s2.0-B9780128117415000199-f19-04-9780128117415.jpg?generation=1733209047672529&alt=media\")
Also, TS_5_4 is a test file for Noise2Void training. No overfitting shenanigans :)
\nThis is a link to the code (it's really messy sry): https://www.kaggle.com/code/kawaiicoderuwu/3d-cell-in-blender
", "messages": [ { "id": 3061972, "postDate": "2024-12-03T07:15:33.797Z", "content": "
I am currently creating 3d reconstructions of the cell scans. My method is as follows:
\n
This is a picture of a cluster of three apo-ferritin in the file TS_5_4 and you can kind of see it, but there is supposed to be a hole as shown below, so it's a little noisy. If anyone wants to try this problem out it's fun :) Also, maybe it can lead to some insights on how to solve the main goal of this competition.
\n
Also, TS_5_4 is a test file for Noise2Void training. No overfitting shenanigans :)
\nThis is a link to the code (it's really messy sry): https://www.kaggle.com/code/kawaiicoderuwu/3d-cell-in-blender
", "rawMarkdown": "\n\nI am currently creating 3d reconstructions of the cell scans. My method is as follows:\n1. Use a Noise2Void model to reduce noise (5 training, 2 test)\n2. Run a 3d weighted filter with a higher weight toward the center\n3. Set all pixels greater than 0.5 standard deviations above the mean as 1 and the rest as 0\n4. Use a 2d fill method to get rid of smaller patches less than 10 pixels to denoise more\n5. Use a 3d fill method to get rid of patches less than 10000 pixels to denoise more\n6. Use marching cubes to make geometry and convert to blender file\n\n\n\n\nThis is a picture of a cluster of three apo-ferritin in the file TS_5_4 and you can kind of see it, but there is supposed to be a hole as shown below, so it's a little noisy. If anyone wants to try this problem out it's fun :) Also, maybe it can lead to some insights on how to solve the main goal of this competition.\n\n\n\n\nAlso, TS_5_4 is a test file for Noise2Void training. No overfitting shenanigans :)\n\nThis is a link to the code (it's really messy sry): [https://www.kaggle.com/code/kawaiicoderuwu/3d-cell-in-blender](https://www.kaggle.com/code/kawaiicoderuwu/3d-cell-in-blender)", "votes": 5 }, { "id": 3062648, "postDate": "2024-12-03T18:56:25.577Z", "content": "There are some existing tools to streamline this kind of visualization for cryo-ET data. ChimeraX with the ArtiaX and Copick plugins is particularly powerful.
\nSee: https://pmc.ncbi.nlm.nih.gov/articles/PMC9667824/
\n@uermel is the author of both of these plugins.
", "rawMarkdown": "There are some existing tools to streamline this kind of visualization for cryo-ET data. ChimeraX with the ArtiaX and Copick plugins is particularly powerful.\n\nSee: https://pmc.ncbi.nlm.nih.gov/articles/PMC9667824/\n\n@uermel is the author of both of these plugins.", "votes": 2, "isDeleted": true } ], "comments": [ { "id": 3062648, "author_name": "", "author_url": "", "post_date": "2024-12-03T18:56:25.577000", "content": "There are some existing tools to streamline this kind of visualization for cryo-ET data. ChimeraX with the ArtiaX and Copick plugins is particularly powerful.
\nSee: https://pmc.ncbi.nlm.nih.gov/articles/PMC9667824/
\n@uermel is the author of both of these plugins.
", "votes": 2, "replies": [] } ], "raw_markdown_by_id": { "3061972": "\n\nI am currently creating 3d reconstructions of the cell scans. My method is as follows:\n1. Use a Noise2Void model to reduce noise (5 training, 2 test)\n2. Run a 3d weighted filter with a higher weight toward the center\n3. Set all pixels greater than 0.5 standard deviations above the mean as 1 and the rest as 0\n4. Use a 2d fill method to get rid of smaller patches less than 10 pixels to denoise more\n5. Use a 3d fill method to get rid of patches less than 10000 pixels to denoise more\n6. Use marching cubes to make geometry and convert to blender file\n\n\n\n\nThis is a picture of a cluster of three apo-ferritin in the file TS_5_4 and you can kind of see it, but there is supposed to be a hole as shown below, so it's a little noisy. If anyone wants to try this problem out it's fun :) Also, maybe it can lead to some insights on how to solve the main goal of this competition.\n\n\n\n\nAlso, TS_5_4 is a test file for Noise2Void training. No overfitting shenanigans :)\n\nThis is a link to the code (it's really messy sry): [https://www.kaggle.com/code/kawaiicoderuwu/3d-cell-in-blender](https://www.kaggle.com/code/kawaiicoderuwu/3d-cell-in-blender)", "3062648": "There are some existing tools to streamline this kind of visualization for cryo-ET data. ChimeraX with the ArtiaX and Copick plugins is particularly powerful.\n\nSee: https://pmc.ncbi.nlm.nih.gov/articles/PMC9667824/\n\n@uermel is the author of both of these plugins." } }