![\"e79de561c\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F876618%2F00ba58c6a06904306dc971d82822a4db%2FScreenshot%20from%202020-11-17%2011-25-33.png?generation=1605619681975072&alt=media\")
First, I want to thank the organizers for putting together this great dataset and competition. I'm quite excited about it.
\nAccording to my faint memories from med school, some annotations seem slightly inaccurate. Image e79de561c seems to have all annotations shifted to the right:
\n
Moreover, some other images appear to have very rare false-positive masks.
\nNote that I could also be wrong as pathology is not my subspecialty (I'm a radiologist).
\nMy comments are meant to help participants in their work and are in no way intended to detract from the HuBMAP project, which will serve as an excellent resource to help foster useful and reproducible AI research.
", "messages": [ { "id": "1081966", "postDate": "11/17/2020 13:28:48", "content": "First, I want to thank the organizers for putting together this great dataset and competition. I'm quite excited about it.
\nAccording to my faint memories from med school, some annotations seem slightly inaccurate. Image e79de561c seems to have all annotations shifted to the right:
\n
Moreover, some other images appear to have very rare false-positive masks.
\nNote that I could also be wrong as pathology is not my subspecialty (I'm a radiologist).
\nMy comments are meant to help participants in their work and are in no way intended to detract from the HuBMAP project, which will serve as an excellent resource to help foster useful and reproducible AI research.
", "rawMarkdown": "First, I want to thank the organizers for putting together this great dataset and competition. I'm quite excited about it.\n\nAccording to my faint memories from med school, some annotations seem slightly inaccurate. Image e79de561c seems to have all annotations shifted to the right:\n\n\n\nMoreover, some other images appear to have very rare false-positive masks.\n\nNote that I could also be wrong as pathology is not my subspecialty (I'm a radiologist).\n\nMy comments are meant to help participants in their work and are in no way intended to detract from the HuBMAP project, which will serve as an excellent resource to help foster useful and reproducible AI research.", "votes": null }, { "id": "1081980", "postDate": "11/17/2020 13:42:29", "content": "I estimate the shift to be 50 pixels to the right.
\n@addisonhoward, are you allowed to disclose if the test set may contain this kind of shift in the annotation? Maybe a more comprehensive question would be: was the annotation process different for the test set (when compared to the annotation for the training set) to guarantee a more accurate ground truth?
", "rawMarkdown": "I estimate the shift to be 50 pixels to the right.\n\n@addisonhoward, are you allowed to disclose if the test set may contain this kind of shift in the annotation? Maybe a more comprehensive question would be: was the annotation process different for the test set (when compared to the annotation for the training set) to guarantee a more accurate ground truth?", "votes": null }, { "id": "1082019", "postDate": "11/17/2020 14:16:18", "content": "It looks like there are also other issues, like inaccurate masks (too rough polygonization?) (0486052bb.tiff)
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F416539%2F0e447c4fe795b097d136e7139be76073%2Fannotation_issue.png?generation=1605622260492931&alt=media\")
That happened to some segmentations too, although not quite often…
\nIt would be interesting to know the annotation process for both training and test sets. That kind of information was already disclosed in other kaggle competitions, like the RSNA Pneumonia Challenge. I think it helps people develop better models.
", "rawMarkdown": "That happened to some segmentations too, although not quite often...\n\nIt would be interesting to know the annotation process for both training and test sets. That kind of information was already disclosed in other kaggle competitions, like the RSNA Pneumonia Challenge. I think it helps people develop better models.", "votes": null }, { "id": "1082167", "postDate": "11/17/2020 16:51:10", "content": "Also, test image c68fe75ea has no information on anatomical structure:
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F232403%2F713e435b74cba771eb704fa3b9ae91f5%2FScreen%20Shot%202020-11-17%20at%2013.51.20.png?generation=1605631862087254&alt=media\")
Another one shifted.
", "rawMarkdown": "Another one shifted.", "votes": null }, { "id": "1082933", "postDate": "11/18/2020 11:52:42", "content": "Is this patch also from image e79de561c? Or is it from a different image?
", "rawMarkdown": "Is this patch also from image e79de561c? Or is it from a different image?", "votes": null }, { "id": "1083058", "postDate": "11/18/2020 14:50:33", "content": "Yes this patch is from the image e79de561c.
", "rawMarkdown": "Yes this patch is from the image e79de561c.", "votes": null }, { "id": "1083068", "postDate": "11/18/2020 14:59:57", "content": "Good, so, what I meant was that the entire e79de561c image is shifted. Any patch from this image will be shifted.
", "rawMarkdown": "Good, so, what I meant was that the entire e79de561c image is shifted. Any patch from this image will be shifted.", "votes": null }, { "id": "1083291", "postDate": "11/18/2020 20:07:29", "content": "We should have the corrected files posted later today. Thank you for the feedback.
", "rawMarkdown": "We should have the corrected files posted later today. Thank you for the feedback.", "votes": null }, { "id": "1083332", "postDate": "11/18/2020 21:46:54", "content": "Thank you, @budsims !
", "rawMarkdown": "Thank you, @budsims !", "votes": null }, { "id": "1083710", "postDate": "11/19/2020 09:18:30", "content": "I think those images are shifted.
\n54f2eec69.tiff (up shift)
\ne79de561c.tiff (right down shift)
\n095bf7a1f.tiff (right down shift)
the data is updated??
", "rawMarkdown": "I think those images are shifted.\n54f2eec69.tiff (up shift)\ne79de561c.tiff (right down shift)\n095bf7a1f.tiff (right down shift)\n\nthe data is updated??", "votes": null }, { "id": "1084167", "postDate": "11/19/2020 19:51:06", "content": "file is corrected now
", "rawMarkdown": "file is corrected now", "votes": null }, { "id": "1084173", "postDate": "11/19/2020 20:02:04", "content": "Thank you!
", "rawMarkdown": "Thank you!", "votes": null }, { "id": "1084224", "postDate": "11/19/2020 20:56:54", "content": "Hi ZZemily,
\ne79de561c has been updated.
\nFor
\n54f2eec69.tiff (up shift)
\n095bf7a1f.tiff (right down shift)
\nThey look correct on my machine. I zoomed in to check to find several glomeruli (not many) that might have 1-2 pixels offset. I think it is limited by the accuracy of the annotation generation but not some offsets on the whole image.
Thank you！！！
", "rawMarkdown": "Thank you！！！", "votes": null } ], "comments": [ { "id": 1081980, "author_name": "felipekitamura", "author_url": "", "post_date": "11/17/2020 13:42:29", "content": "I estimate the shift to be 50 pixels to the right.
\n@addisonhoward, are you allowed to disclose if the test set may contain this kind of shift in the annotation? Maybe a more comprehensive question would be: was the annotation process different for the test set (when compared to the annotation for the training set) to guarantee a more accurate ground truth?
", "votes": null, "replies": [] }, { "id": 1082019, "author_name": "stpise", "author_url": "", "post_date": "11/17/2020 14:16:18", "content": "It looks like there are also other issues, like inaccurate masks (too rough polygonization?) (0486052bb.tiff)
\n
That happened to some segmentations too, although not quite often…
\nIt would be interesting to know the annotation process for both training and test sets. That kind of information was already disclosed in other kaggle competitions, like the RSNA Pneumonia Challenge. I think it helps people develop better models.
", "votes": null, "replies": [] } ] }, { "id": 1082167, "author_name": "bguberfain", "author_url": "", "post_date": "11/17/2020 16:51:10", "content": "Also, test image c68fe75ea has no information on anatomical structure:
\n
file is corrected now
", "votes": null, "replies": [] }, { "id": 1084173, "author_name": "felipekitamura", "author_url": "", "post_date": "11/19/2020 20:02:04", "content": "Thank you!
", "votes": null, "replies": [] } ] }, { "id": 1082765, "author_name": "nxrprime", "author_url": "", "post_date": "11/18/2020 07:54:06", "content": "Another one shifted.
", "votes": null, "replies": [ { "id": 1082933, "author_name": "felipekitamura", "author_url": "", "post_date": "11/18/2020 11:52:42", "content": "Is this patch also from image e79de561c? Or is it from a different image?
", "votes": null, "replies": [] }, { "id": 1083058, "author_name": "nxrprime", "author_url": "", "post_date": "11/18/2020 14:50:33", "content": "Yes this patch is from the image e79de561c.
", "votes": null, "replies": [] }, { "id": 1083068, "author_name": "felipekitamura", "author_url": "", "post_date": "11/18/2020 14:59:57", "content": "Good, so, what I meant was that the entire e79de561c image is shifted. Any patch from this image will be shifted.
", "votes": null, "replies": [] } ] }, { "id": 1083291, "author_name": "budsims", "author_url": "", "post_date": "11/18/2020 20:07:29", "content": "We should have the corrected files posted later today. Thank you for the feedback.
", "votes": null, "replies": [ { "id": 1083332, "author_name": "felipekitamura", "author_url": "", "post_date": "11/18/2020 21:46:54", "content": "Thank you, @budsims !
", "votes": null, "replies": [] } ] }, { "id": 1083710, "author_name": "zzemily", "author_url": "", "post_date": "11/19/2020 09:18:30", "content": "I think those images are shifted.
\n54f2eec69.tiff (up shift)
\ne79de561c.tiff (right down shift)
\n095bf7a1f.tiff (right down shift)
the data is updated??
", "votes": null, "replies": [ { "id": 1084224, "author_name": "juyingnan", "author_url": "", "post_date": "11/19/2020 20:56:54", "content": "Hi ZZemily,
\ne79de561c has been updated.
\nFor
\n54f2eec69.tiff (up shift)
\n095bf7a1f.tiff (right down shift)
\nThey look correct on my machine. I zoomed in to check to find several glomeruli (not many) that might have 1-2 pixels offset. I think it is limited by the accuracy of the annotation generation but not some offsets on the whole image.
Thank you！！！
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1081966": "First, I want to thank the organizers for putting together this great dataset and competition. I'm quite excited about it.\n\nAccording to my faint memories from med school, some annotations seem slightly inaccurate. Image e79de561c seems to have all annotations shifted to the right:\n\n\n\nMoreover, some other images appear to have very rare false-positive masks.\n\nNote that I could also be wrong as pathology is not my subspecialty (I'm a radiologist).\n\nMy comments are meant to help participants in their work and are in no way intended to detract from the HuBMAP project, which will serve as an excellent resource to help foster useful and reproducible AI research.", "1081980": "I estimate the shift to be 50 pixels to the right.\n\n@addisonhoward, are you allowed to disclose if the test set may contain this kind of shift in the annotation? Maybe a more comprehensive question would be: was the annotation process different for the test set (when compared to the annotation for the training set) to guarantee a more accurate ground truth?", "1082019": "It looks like there are also other issues, like inaccurate masks (too rough polygonization?) (0486052bb.tiff)\n\n", "1082036": "That happened to some segmentations too, although not quite often...\n\nIt would be interesting to know the annotation process for both training and test sets. That kind of information was already disclosed in other kaggle competitions, like the RSNA Pneumonia Challenge. I think it helps people develop better models.", "1082167": "Also, test image c68fe75ea has no information on anatomical structure:\n\n", "1082765": "Another one shifted.", "1082933": "Is this patch also from image e79de561c? Or is it from a different image?", "1083058": "Yes this patch is from the image e79de561c.", "1083068": "Good, so, what I meant was that the entire e79de561c image is shifted. Any patch from this image will be shifted.", "1083291": "We should have the corrected files posted later today. Thank you for the feedback.", "1083332": "Thank you, @budsims !", "1083710": "I think those images are shifted.\n54f2eec69.tiff (up shift)\ne79de561c.tiff (right down shift)\n095bf7a1f.tiff (right down shift)\n\nthe data is updated??", "1084167": "file is corrected now", "1084173": "Thank you!", "1084224": "Hi ZZemily,\ne79de561c has been updated.\nFor\n54f2eec69.tiff (up shift)\n095bf7a1f.tiff (right down shift)\nThey look correct on my machine. I zoomed in to check to find several glomeruli (not many) that might have 1-2 pixels offset. I think it is limited by the accuracy of the annotation generation but not some offsets on the whole image.", "1084375": "Thank you！！！" }, "source": "meta" }