{ "id": 341732, "title": "Have lungs been segmented with deep learning?", "url": "/competitions/hubmap-organ-segmentation/discussion/341732", "author_name": "", "post_date": "2022-08-04T05:01:19.094396600Z", "votes": 7, "comment_count": 3, "views": 0, "content": "

I see very strange-looking segmentation masks for lungs. Wonder if those masks were received with some automatic segmentation process. Are these really what we are looking for? Are these just bad masks?
\n\"\"

\n

\"\"

\n

\"\"

", "messages": [ { "id": "1883766", "postDate": "08/04/2022 05:01:19", "content": "

I see very strange-looking segmentation masks for lungs. Wonder if those masks were received with some automatic segmentation process. Are these really what we are looking for? Are these just bad masks?
\n\"\"

\n

\"\"

\n

\"\"

", "rawMarkdown": "I see very strange-looking segmentation masks for lungs. Wonder if those masks were received with some automatic segmentation process. Are these really what we are looking for? Are these just bad masks?\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F349155%2F941ee9773fbbdf78f32e7d49b9dd6c27%2Flungs1.png?generation=1659589313975768&alt=media)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F349155%2F59616a9ddb8c23817dc94aa8628aad0f%2Flungs4.png?generation=1659589388416628&alt=media)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F349155%2F108e08b673a3ed762746a7e455e0a38d%2Flungs3.png?generation=1659589338686388&alt=media)", "votes": null }, { "id": "1887944", "postDate": "08/07/2022 07:50:34", "content": "

The lungs seem to be the most difficult to segment. My mean IoU CV scores for each organ is ~0.70, the lung being the exception with a mean IoU of ~0.10…

", "rawMarkdown": "The lungs seem to be the most difficult to segment. My mean IoU CV scores for each organ is ~0.70, the lung being the exception with a mean IoU of ~0.10...", "votes": null }, { "id": "1887951", "postDate": "08/07/2022 08:06:30", "content": "

the answer is yes. but for very little of them. it is shown by the lb-hubmap score. though the score is low, it is better than random guess or just submitting some fix region like central circle.

\n

you can also plot the prediction heat map of your validation set or other lung images from the internet. the model did learn something

", "rawMarkdown": "the answer is yes. but for very little of them. it is shown by the lb-hubmap score. though the score is low, it is better than random guess or just submitting some fix region like central circle.\n\nyou can also plot the prediction heat map of your validation set or other lung images from the internet. the model did learn something", "votes": null }, { "id": "1910359", "postDate": "08/23/2022 11:17:50", "content": "

I'm still trying to figure out how lungs are annotated. It doesn't look like they are model outputs because of the sharp edges. Annotators were definitely using polygons.

", "rawMarkdown": "I'm still trying to figure out how lungs are annotated. It doesn't look like they are model outputs because of the sharp edges. Annotators were definitely using polygons.", "votes": null } ], "comments": [ { "id": 1887944, "author_name": "markwijkhuizen", "author_url": "", "post_date": "08/07/2022 07:50:34", "content": "

The lungs seem to be the most difficult to segment. My mean IoU CV scores for each organ is ~0.70, the lung being the exception with a mean IoU of ~0.10…

", "votes": null, "replies": [] }, { "id": 1887951, "author_name": "hengck23", "author_url": "", "post_date": "08/07/2022 08:06:30", "content": "

the answer is yes. but for very little of them. it is shown by the lb-hubmap score. though the score is low, it is better than random guess or just submitting some fix region like central circle.

\n

you can also plot the prediction heat map of your validation set or other lung images from the internet. the model did learn something

", "votes": null, "replies": [] }, { "id": 1910359, "author_name": "gunesevitan", "author_url": "", "post_date": "08/23/2022 11:17:50", "content": "

I'm still trying to figure out how lungs are annotated. It doesn't look like they are model outputs because of the sharp edges. Annotators were definitely using polygons.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1883766": "I see very strange-looking segmentation masks for lungs. Wonder if those masks were received with some automatic segmentation process. Are these really what we are looking for? Are these just bad masks?\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F349155%2F941ee9773fbbdf78f32e7d49b9dd6c27%2Flungs1.png?generation=1659589313975768&alt=media)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F349155%2F59616a9ddb8c23817dc94aa8628aad0f%2Flungs4.png?generation=1659589388416628&alt=media)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F349155%2F108e08b673a3ed762746a7e455e0a38d%2Flungs3.png?generation=1659589338686388&alt=media)", "1887944": "The lungs seem to be the most difficult to segment. My mean IoU CV scores for each organ is ~0.70, the lung being the exception with a mean IoU of ~0.10...", "1887951": "the answer is yes. but for very little of them. it is shown by the lb-hubmap score. though the score is low, it is better than random guess or just submitting some fix region like central circle.\n\nyou can also plot the prediction heat map of your validation set or other lung images from the internet. the model did learn something", "1910359": "I'm still trying to figure out how lungs are annotated. It doesn't look like they are model outputs because of the sharp edges. Annotators were definitely using polygons." }, "source": "meta" }