\n
![\"Plot](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F24737244%2Fb6274462474aae7a81636f90a2f25eff%2FMABe22_movies_1100085542_1100.png?generation=1759781412663548&alt=media\")
\nCode to reproduce, with some examples of other labs with plausible mouse/bodypart allocations, including MABe22_keypoints which does seem to be fine, in this notebook.
I'm posting a data patch now that should address some of the issues raised in various threads throughout the forums. It should be available for download soon. We'll post a more detailed description of the patch when possible.
", "messages": [ { "id": 3294840, "postDate": "2025-09-26T22:14:48.123Z", "content": "I'm posting a data patch now that should address some of the issues raised in various threads throughout the forums. It should be available for download soon. We'll post a more detailed description of the patch when possible.
", "rawMarkdown": "I'm posting a data patch now that should address some of the issues raised in various threads throughout the forums. It should be available for download soon. We'll post a more detailed description of the patch when possible.", "votes": 22 }, { "id": 3295071, "postDate": "2025-09-27T15:24:05.977Z", "content": "Thank you all for your investigation into the dataset that helped identify these issues.
\nChanges in this patch:
\nThank you for the data patch notes.
\nAccording to the notes, the \"ejaculate\" label was supposed to be removed. However, after downloading the latest version of the dataset using the \"Download All\" button and re-tallying train.csv, it appears that three instances of the \"ejaculate\" label still exist.
\nI also confirmed that the annotation data for the following IDs still contain the \"ejaculate\" label: 551378303, 1176870705, and 1715945836.
\nIs this intended behavior, or could this be an issue on my end? I would appreciate it if you could look into this.
", "rawMarkdown": "Thank you for the data patch notes.\n\nAccording to the notes, the \"ejaculate\" label was supposed to be removed. However, after downloading the latest version of the dataset using the \"Download All\" button and re-tallying train.csv, it appears that three instances of the \"ejaculate\" label still exist.\n\nI also confirmed that the annotation data for the following IDs still contain the \"ejaculate\" label: 551378303, 1176870705, and 1715945836.\n\nIs this intended behavior, or could this be an issue on my end? I would appreciate it if you could look into this.", "votes": 2, "replies": [ { "id": 3295340, "postDate": "2025-09-28T13:18:58.603Z", "content": "Thanks for catching that, I will remove it from the notes- we have gone back and forth on whether to keep it. It does not appear in the test set so can be safely ignored.
", "rawMarkdown": "Thanks for catching that, I will remove it from the notes- we have gone back and forth on whether to keep it. It does not appear in the test set so can be safely ignored.", "votes": 4 } ] }, { "id": 3295446, "postDate": "2025-09-28T19:19:32.110Z", "content": "Also SparklingTapir now miss 15 files with annotations
", "rawMarkdown": "Also SparklingTapir now miss 15 files with annotations", "replies": [ { "id": 3295764, "postDate": "2025-09-29T13:51:41.337Z", "content": "Yes- these are trials that previously only had annotations for sniffgenital and chase. After removing these labels there were no scored actions in the videos, and hence no annotation files. We still include these data as we have found that self-supervised representation learning of animal movement is often a useful first step for behavior classification.
", "rawMarkdown": "Yes- these are trials that previously only had annotations for sniffgenital and chase. After removing these labels there were no scored actions in the videos, and hence no annotation files. We still include these data as we have found that self-supervised representation learning of animal movement is often a useful first step for behavior classification.", "votes": 4 } ] }, { "id": 3298960, "postDate": "2025-10-06T20:20:30.980Z", "content": "@annkennedy05, @sohier, I regret to report that the mouse/bodypart allocation of keypoints in MABe22_movies is not fixed in the latest dataset version!
\n
\nCode to reproduce, with some examples of other labs with plausible mouse/bodypart allocations, including MABe22_keypoints which does seem to be fine, in this notebook.
Thank you all for your investigation into the dataset.
", "rawMarkdown": "Thank you all for your investigation into the dataset." }, { "id": 3296285, "postDate": "2025-09-30T15:12:25.783Z", "content": "Whenever I am generating submission.csv file I am only getting one entries. And also in my sample_submission.csv has only one entries and test tracking has only one perquite file. How should I create more output for submission.csv file.
", "rawMarkdown": "Whenever I am generating submission.csv file I am only getting one entries. And also in my sample_submission.csv has only one entries and test tracking has only one perquite file. How should I create more output for submission.csv file." }, { "id": 3295079, "postDate": "2025-09-27T15:34:41.113Z", "content": "Hey!
\nNot sure if reply is the best way to tackle this - for video 1260392287 (AdaptableSnail) - the csv defines data for mouse1, mouse2, mouse3 (mouse4 data is nan). Tracking data in \"dataset/train_tracking/AdaptableSnail/1260392287.parquet\" contains parts coordinates for mouse1, mouse2 & mouse4. CSV behaviors_labeled is even strager as it contains all 4 mouse1, mouse2, mouse3 & mouse4.
\nPlus 2 annotation files missing: 1350372183 & 1375833299
\nThanks
", "rawMarkdown": "Hey! \n\nNot sure if reply is the best way to tackle this - for video 1260392287 (AdaptableSnail) - the csv defines data for mouse1, mouse2, mouse3 (mouse4 data is nan). Tracking data in \"dataset/train_tracking/AdaptableSnail/1260392287.parquet\" contains parts coordinates for mouse1, mouse2 & mouse4. CSV behaviors_labeled is even strager as it contains all 4 mouse1, mouse2, mouse3 & mouse4. \n\nPlus 2 annotation files missing: 1350372183 & 1375833299\n\nThanks", "replies": [ { "id": 3295774, "postDate": "2025-09-29T14:16:08.907Z", "content": "Thank you for bringing this to our attention! The files in train_tracking and train_annotation for this video both use mouse 1, 2, and 4. This is an artifact of how mice were ID'd in the dataset- by a colored marking, which we mapped to mouse numbers. Sometimes the mouse 3 color wasn't present but the mouse 4 color was. We will update the datasheet so the populated columns match the files.
\nThe csv behaviors_labeled column is more troubling, and doesn't appear in my local copy- will look into this asap.
\nRegarding missing annotation files, these are videos in which annotators looked for behaviors but did not observe them, thus there were no annotations (see my reply to Antonina above.)
", "rawMarkdown": "Thank you for bringing this to our attention! The files in train_tracking and train_annotation for this video both use mouse 1, 2, and 4. This is an artifact of how mice were ID'd in the dataset- by a colored marking, which we mapped to mouse numbers. Sometimes the mouse 3 color wasn't present but the mouse 4 color was. We will update the datasheet so the populated columns match the files.\n\nThe csv behaviors_labeled column is more troubling, and doesn't appear in my local copy- will look into this asap.\n\nRegarding missing annotation files, these are videos in which annotators looked for behaviors but did not observe them, thus there were no annotations (see my reply to Antonina above.)" } ] }, { "id": 3295193, "postDate": "2025-09-28T01:57:28.960Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 3295071, "author_name": "Ann Kennedy", "author_url": "", "post_date": "2025-09-27T15:24:05.977000", "content": "Thank you all for your investigation into the dataset that helped identify these issues.
\nChanges in this patch:
\nThank you for the data patch notes.
\nAccording to the notes, the \"ejaculate\" label was supposed to be removed. However, after downloading the latest version of the dataset using the \"Download All\" button and re-tallying train.csv, it appears that three instances of the \"ejaculate\" label still exist.
\nI also confirmed that the annotation data for the following IDs still contain the \"ejaculate\" label: 551378303, 1176870705, and 1715945836.
\nIs this intended behavior, or could this be an issue on my end? I would appreciate it if you could look into this.
", "votes": 2, "replies": [ { "id": 3295340, "author_name": "Ann Kennedy", "author_url": "", "post_date": "2025-09-28T13:18:58.603000", "content": "Thanks for catching that, I will remove it from the notes- we have gone back and forth on whether to keep it. It does not appear in the test set so can be safely ignored.
", "votes": 4, "replies": [] } ] }, { "id": 3295446, "author_name": "Antonina Dolgorukova", "author_url": "", "post_date": "2025-09-28T19:19:32.110000", "content": "Also SparklingTapir now miss 15 files with annotations
", "votes": 0, "replies": [ { "id": 3295764, "author_name": "Ann Kennedy", "author_url": "", "post_date": "2025-09-29T13:51:41.337000", "content": "Yes- these are trials that previously only had annotations for sniffgenital and chase. After removing these labels there were no scored actions in the videos, and hence no annotation files. We still include these data as we have found that self-supervised representation learning of animal movement is often a useful first step for behavior classification.
", "votes": 4, "replies": [] } ] }, { "id": 3298960, "author_name": "Will Hirsch", "author_url": "", "post_date": "2025-10-06T20:20:30.980000", "content": "@annkennedy05, @sohier, I regret to report that the mouse/bodypart allocation of keypoints in MABe22_movies is not fixed in the latest dataset version!
\n
\nCode to reproduce, with some examples of other labs with plausible mouse/bodypart allocations, including MABe22_keypoints which does seem to be fine, in this notebook.
Thank you all for your investigation into the dataset.
", "votes": 0, "replies": [] }, { "id": 3296285, "author_name": "Pawan26 Singh", "author_url": "", "post_date": "2025-09-30T15:12:25.783000", "content": "Whenever I am generating submission.csv file I am only getting one entries. And also in my sample_submission.csv has only one entries and test tracking has only one perquite file. How should I create more output for submission.csv file.
", "votes": 0, "replies": [] }, { "id": 3295079, "author_name": "vinpro", "author_url": "", "post_date": "2025-09-27T15:34:41.113000", "content": "Hey!
\nNot sure if reply is the best way to tackle this - for video 1260392287 (AdaptableSnail) - the csv defines data for mouse1, mouse2, mouse3 (mouse4 data is nan). Tracking data in \"dataset/train_tracking/AdaptableSnail/1260392287.parquet\" contains parts coordinates for mouse1, mouse2 & mouse4. CSV behaviors_labeled is even strager as it contains all 4 mouse1, mouse2, mouse3 & mouse4.
\nPlus 2 annotation files missing: 1350372183 & 1375833299
\nThanks
", "votes": 0, "replies": [ { "id": 3295774, "author_name": "Ann Kennedy", "author_url": "", "post_date": "2025-09-29T14:16:08.907000", "content": "Thank you for bringing this to our attention! The files in train_tracking and train_annotation for this video both use mouse 1, 2, and 4. This is an artifact of how mice were ID'd in the dataset- by a colored marking, which we mapped to mouse numbers. Sometimes the mouse 3 color wasn't present but the mouse 4 color was. We will update the datasheet so the populated columns match the files.
\nThe csv behaviors_labeled column is more troubling, and doesn't appear in my local copy- will look into this asap.
\nRegarding missing annotation files, these are videos in which annotators looked for behaviors but did not observe them, thus there were no annotations (see my reply to Antonina above.)
", "votes": 0, "replies": [] } ] }, { "id": 3295193, "author_name": "", "author_url": "", "post_date": "2025-09-28T01:57:28.960000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3294840": "I'm posting a data patch now that should address some of the issues raised in various threads throughout the forums. It should be available for download soon. We'll post a more detailed description of the patch when possible.", "3295071": "Thank you all for your investigation into the dataset that helped identify these issues.\n\nChanges in this patch:\n- **CautiousGiraffe**: corrected arena size metadata from 50x50cm to 30x30cm (impacts both training + test data.)\n- **MABe22_movies**: undid a previous rescaling of tracking data so it now aligns with 800x800px arena dimensions listed in metadata (impacts training data only.)\n- **MABe22_keypoints** and **MABe22_movies**: corrected keypoint formatting error that swapped some body parts between mice (impacts training data only.)\n- **SparklingTapir**: removed one video where tracking data didn’t match expected video dimensions. Corrected pix/cm resolution data for six videos. Removed two low-quality behavior labels (sniffgenital and chase) that do not appear in test set. (impacts training data only.)", "3296366": "Thank you all for your investigation into the dataset.", "3296285": "Whenever I am generating submission.csv file I am only getting one entries. And also in my sample_submission.csv has only one entries and test tracking has only one perquite file. How should I create more output for submission.csv file.", "3295079": "Hey! \n\nNot sure if reply is the best way to tackle this - for video 1260392287 (AdaptableSnail) - the csv defines data for mouse1, mouse2, mouse3 (mouse4 data is nan). Tracking data in \"dataset/train_tracking/AdaptableSnail/1260392287.parquet\" contains parts coordinates for mouse1, mouse2 & mouse4. CSV behaviors_labeled is even strager as it contains all 4 mouse1, mouse2, mouse3 & mouse4. \n\nPlus 2 annotation files missing: 1350372183 & 1375833299\n\nThanks", "3295193": "" } }