![\"\"](\"https://raw.githubusercontent.com/wasserth/TotalSegmentator/master/resources/imgs/overview_classes.png\")
https://github.com/wasserth/TotalSegmentator, from RSNA, which can split a lot of content, and I'm looking into whether it can be used directly for this competition.
\n
![\"\"](\"https://raw.githubusercontent.com/wasserth/TotalSegmentator/master/resources/imgs/overview_subclasses.png\")
In fact, we can use it directly! In the figure below, on the left is the semantic segmentation label given by the game, and on the right is the prediction of the TotalSegmentator. It works very well! (A little slow, but the patience is worth it!)
\n![\"1\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8314415%2F9533bb74e37629b551e3e1de327ea36a%2Fanimation_pred1.gif?generation=1691052188062592&alt=media\")
Now the only thing I want to know is if we can just use this pre-trained model?🤔
", "messages": [ { "id": 2371685, "postDate": "2023-08-03T08:40:38.580Z", "content": "https://github.com/wasserth/TotalSegmentator, from RSNA, which can split a lot of content, and I'm looking into whether it can be used directly for this competition.
\n
In fact, we can use it directly! In the figure below, on the left is the semantic segmentation label given by the game, and on the right is the prediction of the TotalSegmentator. It works very well! (A little slow, but the patience is worth it!)
\nNow the only thing I want to know is if we can just use this pre-trained model?🤔
", "rawMarkdown": "[https://github.com/wasserth/TotalSegmentator]( https://github.com/wasserth/TotalSegmentator), from RSNA, which can split a lot of content, and I'm looking into whether it can be used directly for this competition. \n\n\n\n\n\n---\n\n**In fact, we can use it directly! In the figure below, on the left is the semantic segmentation label given by the game, and on the right is the prediction of the TotalSegmentator. It works very well! (A little slow, but the patience is worth it!)**\n\n\n---\n\nNow the only thing I want to know is if we can just use this pre-trained model?🤔", "votes": 11 }, { "id": 2372824, "postDate": "2023-08-04T01:54:26.420Z", "content": "@davidbroberts is right. but unfortunately, after my manual review, I realized that TotalSegmentor's categories may be a little less identical to the contest requirements.
\nAs you can see, TotalSegmentor's does not contain class extravasation
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8314415%2F53defecc92a275c96f6c5f21e1d7b53d%2F5e6cc4971a13bd66dccebddda2f2abd.png?generation=1691113961213388&alt=media\")
You can check gif there
", "rawMarkdown": "@davidbroberts is right. but unfortunately, after my manual review, I realized that TotalSegmentor's categories may be a little less identical to the contest requirements.\n\nAs you can see, TotalSegmentor's does not contain class `extravasation`\n\n\n\nYou can check gif [there](https://drive.google.com/file/d/1_6QXzwoLg2aNezjNN94Xaq2gXEhLjxsj/view?usp=sharing)", "votes": 3, "replies": [ { "id": 2374296, "postDate": "2023-08-04T20:35:51.770Z", "content": "According to my understanding, extravasation refers to a state where active bleeding is persisting, so the region indicated by the mask may not directly represent it. Without knowing the original image, it's uncertain, but I believe the mask is showing the hematoma resulting from extravasation.
\nBy the way, thank you for sharing the excellent GIF. Despite the considerable hematoma, it seems that other organs are relatively well segmented. I wonder if totalsegmentator could be very useful in this competition depending on how it's used.
", "rawMarkdown": "According to my understanding, extravasation refers to a state where active bleeding is persisting, so the region indicated by the mask may not directly represent it. Without knowing the original image, it's uncertain, but I believe the mask is showing the hematoma resulting from extravasation.\n\nBy the way, thank you for sharing the excellent GIF. Despite the considerable hematoma, it seems that other organs are relatively well segmented. I wonder if totalsegmentator could be very useful in this competition depending on how it's used.", "votes": 3, "replies": [ { "id": 2427287, "postDate": "2023-09-07T06:30:01.353Z", "content": "Where is the hematoma? I can not see it very well. Is it segmented as some other organ, or ignored by the segmentator?
", "rawMarkdown": "Where is the hematoma? I can not see it very well. Is it segmented as some other organ, or ignored by the segmentator?" } ] } ] }, { "id": 2372039, "postDate": "2023-08-03T12:46:16.250Z", "content": "TotalSegmentor is the tool the dataset was created with.
", "rawMarkdown": "TotalSegmentor is the tool the dataset was created with.", "votes": 2 }, { "id": 2427303, "postDate": "2023-09-07T06:37:38.193Z", "content": "Why are the colours different? I think it makes it harder to see whether everything is matching up….
", "rawMarkdown": "Why are the colours different? I think it makes it harder to see whether everything is matching up....", "votes": -1 }, { "id": 2398683, "postDate": "2023-08-19T20:35:29.787Z", "content": "I'm a bit confused on whether this means that we can use TotalSegmentator or not in our solution or whether we need to train our own segmentation model?
", "rawMarkdown": "I'm a bit confused on whether this means that we can use TotalSegmentator or not in our solution or whether we need to train our own segmentation model? ", "replies": [ { "id": 2399080, "postDate": "2023-08-20T06:11:20.163Z", "content": "Assuming you can get it running on your compute you can use it to train. I am having a bit of trouble getting it to run on my Ubuntu 22.04.03 system but have not tried real hard to work out the issue.
\nDo you need a segmentation model depends on your approach. I don't think you need one if your doing a 3D model, but pretty sure to get any score better than the 'means' you will need to segment for a 2D model.
", "rawMarkdown": "Assuming you can get it running on your compute you can use it to train. I am having a bit of trouble getting it to run on my Ubuntu 22.04.03 system but have not tried real hard to work out the issue.\n\nDo you need a segmentation model depends on your approach. I don't think you need one if your doing a 3D model, but pretty sure to get any score better than the 'means' you will need to segment for a 2D model.", "replies": [ { "id": 2427301, "postDate": "2023-09-07T06:36:34.287Z", "content": "Why is segmentation not useful when using 3D model?
", "rawMarkdown": "Why is segmentation not useful when using 3D model?" } ] } ] }, { "id": 2372017, "postDate": "2023-08-03T12:26:05.947Z", "content": "I think that is not accepted. In Code Requirements You can see 'Internet access disabled'. You just use libraries/frameworks contained in kaggle notebook. You can not install other packages like TotalSegmentator
", "rawMarkdown": "I think that is not accepted. In Code Requirements You can see 'Internet access disabled'. You just use libraries/frameworks contained in kaggle notebook. You can not install other packages like TotalSegmentator", "replies": [ { "id": 2372048, "postDate": "2023-08-03T12:47:52.467Z", "content": "You can install other libraries into Kaggle during \"internet-off\" competitions. You load them as datasets. I don't know specifically about this package but other packages can be used.
", "rawMarkdown": "You can install other libraries into Kaggle during \"internet-off\" competitions. You load them as datasets. I don't know specifically about this package but other packages can be used.", "votes": 1, "replies": [ { "id": 2372287, "postDate": "2023-08-03T15:41:03.833Z", "content": "ok, thank you, i wil try.
", "rawMarkdown": "ok, thank you, i wil try." }, { "id": 2413517, "postDate": "2023-08-29T01:15:29.240Z", "content": "Assuming the model learns from seperated organs, with everything else black, the model should already have some segmentation ability. Right? Since it should know to \"focus\" on the segmented area.
", "rawMarkdown": "Assuming the model learns from seperated organs, with everything else black, the model should already have some segmentation ability. Right? Since it should know to \"focus\" on the segmented area." } ] } ] } ], "comments": [ { "id": 2372824, "author_name": "JavaZero", "author_url": "", "post_date": "2023-08-04T01:54:26.420000", "content": "@davidbroberts is right. but unfortunately, after my manual review, I realized that TotalSegmentor's categories may be a little less identical to the contest requirements.
\nAs you can see, TotalSegmentor's does not contain class extravasation
You can check gif there
", "votes": 3, "replies": [ { "id": 2374296, "author_name": "YYama", "author_url": "", "post_date": "2023-08-04T20:35:51.770000", "content": "According to my understanding, extravasation refers to a state where active bleeding is persisting, so the region indicated by the mask may not directly represent it. Without knowing the original image, it's uncertain, but I believe the mask is showing the hematoma resulting from extravasation.
\nBy the way, thank you for sharing the excellent GIF. Despite the considerable hematoma, it seems that other organs are relatively well segmented. I wonder if totalsegmentator could be very useful in this competition depending on how it's used.
", "votes": 3, "replies": [ { "id": 2427287, "author_name": "Gyula Maloveczky4", "author_url": "", "post_date": "2023-09-07T06:30:01.353000", "content": "Where is the hematoma? I can not see it very well. Is it segmented as some other organ, or ignored by the segmentator?
", "votes": 0, "replies": [] } ] } ] }, { "id": 2372039, "author_name": "David Roberts", "author_url": "", "post_date": "2023-08-03T12:46:16.250000", "content": "TotalSegmentor is the tool the dataset was created with.
", "votes": 2, "replies": [] }, { "id": 2427303, "author_name": "Gyula Maloveczky4", "author_url": "", "post_date": "2023-09-07T06:37:38.193000", "content": "Why are the colours different? I think it makes it harder to see whether everything is matching up….
", "votes": -1, "replies": [] }, { "id": 2398683, "author_name": "Sam Lee", "author_url": "", "post_date": "2023-08-19T20:35:29.787000", "content": "I'm a bit confused on whether this means that we can use TotalSegmentator or not in our solution or whether we need to train our own segmentation model?
", "votes": 0, "replies": [ { "id": 2399080, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2023-08-20T06:11:20.163000", "content": "Assuming you can get it running on your compute you can use it to train. I am having a bit of trouble getting it to run on my Ubuntu 22.04.03 system but have not tried real hard to work out the issue.
\nDo you need a segmentation model depends on your approach. I don't think you need one if your doing a 3D model, but pretty sure to get any score better than the 'means' you will need to segment for a 2D model.
", "votes": 0, "replies": [ { "id": 2427301, "author_name": "Gyula Maloveczky4", "author_url": "", "post_date": "2023-09-07T06:36:34.287000", "content": "Why is segmentation not useful when using 3D model?
", "votes": 0, "replies": [] } ] } ] }, { "id": 2372017, "author_name": "Vietnamese-NHNAM", "author_url": "", "post_date": "2023-08-03T12:26:05.947000", "content": "I think that is not accepted. In Code Requirements You can see 'Internet access disabled'. You just use libraries/frameworks contained in kaggle notebook. You can not install other packages like TotalSegmentator
", "votes": 0, "replies": [ { "id": 2372048, "author_name": "David Roberts", "author_url": "", "post_date": "2023-08-03T12:47:52.467000", "content": "You can install other libraries into Kaggle during \"internet-off\" competitions. You load them as datasets. I don't know specifically about this package but other packages can be used.
", "votes": 1, "replies": [ { "id": 2372287, "author_name": "Vietnamese-NHNAM", "author_url": "", "post_date": "2023-08-03T15:41:03.833000", "content": "ok, thank you, i wil try.
", "votes": 0, "replies": [] }, { "id": 2413517, "author_name": "LLLEEEOOOH", "author_url": "", "post_date": "2023-08-29T01:15:29.240000", "content": "Assuming the model learns from seperated organs, with everything else black, the model should already have some segmentation ability. Right? Since it should know to \"focus\" on the segmented area.
", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2371685": "[https://github.com/wasserth/TotalSegmentator]( https://github.com/wasserth/TotalSegmentator), from RSNA, which can split a lot of content, and I'm looking into whether it can be used directly for this competition. \n\n\n\n\n\n---\n\n**In fact, we can use it directly! In the figure below, on the left is the semantic segmentation label given by the game, and on the right is the prediction of the TotalSegmentator. It works very well! (A little slow, but the patience is worth it!)**\n\n\n---\n\nNow the only thing I want to know is if we can just use this pre-trained model?🤔", "2372824": "@davidbroberts is right. but unfortunately, after my manual review, I realized that TotalSegmentor's categories may be a little less identical to the contest requirements.\n\nAs you can see, TotalSegmentor's does not contain class `extravasation`\n\n\n\nYou can check gif [there](https://drive.google.com/file/d/1_6QXzwoLg2aNezjNN94Xaq2gXEhLjxsj/view?usp=sharing)", "2372039": "TotalSegmentor is the tool the dataset was created with.", "2427303": "Why are the colours different? I think it makes it harder to see whether everything is matching up....", "2398683": "I'm a bit confused on whether this means that we can use TotalSegmentator or not in our solution or whether we need to train our own segmentation model? ", "2372017": "I think that is not accepted. In Code Requirements You can see 'Internet access disabled'. You just use libraries/frameworks contained in kaggle notebook. You can not install other packages like TotalSegmentator" } }