![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8162991%2F050f1d6a2a5d5524c76e8e4110db19b8%2F1.png?generation=1757764016072101&alt=media\")
Yesterday I ran the training code that worked well on the data before the update, and I saw that on the updated data the exact same code (without any changes) performs terribly.
\nHas anyone else encountered a similar issue?
\nI tried to figure out what happened and noticed that the .dcm files seem to have been renamed, but the values in the SOPInstanceUID column of train_localizers.csv have essentially not changed. Seems like the SOPInstanceUID column in train_localizers.csv is pointing to old dcm filenames (before Data Update 2).
\nBelow you can see my dicts (which I use to sort the .dcm files) InstanceNumber: filename.dcm for one series before and after the update:
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8162991%2Feaa34cccf734cf37c3c59e3147371c40%2F2.png?generation=1757764023178149&alt=media\")
Also try running one of the older visualization notebooks: The output of one of the cells where train_localizers.csv is used has changed
\nhttps://www.kaggle.com/code/perseus7/rsna-exploratory-data-analysis-and-visualization/
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8162991%2Fbc9a015fe372141f9b750f2b6f825b68%2F215.png?generation=1757767116677836&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8162991%2Ffbd402c28a128aef8ceb064b82006bf2%2F214.png?generation=1757767041582775&alt=media\")
UPDATE 1:
\nI assumed that the SOPInstanceUID column in train_localizers.csv (version 2025-09-13, after Data Update 2) still points to the old .dcm filenames (before Data Update 2). By using the saved InstanceNumber values corresponding to the old filenames, I was able to match them with the new .dcm filenames for the same InstanceNumber. After that, my training code started working as before. And the problem was solved.
", "messages": [ { "id": 3288218, "postDate": "2025-09-13T11:48:45.960Z", "content": "Yesterday I ran the training code that worked well on the data before the update, and I saw that on the updated data the exact same code (without any changes) performs terribly.
\nHas anyone else encountered a similar issue?
\nI tried to figure out what happened and noticed that the .dcm files seem to have been renamed, but the values in the SOPInstanceUID column of train_localizers.csv have essentially not changed. Seems like the SOPInstanceUID column in train_localizers.csv is pointing to old dcm filenames (before Data Update 2).
\nBelow you can see my dicts (which I use to sort the .dcm files) InstanceNumber: filename.dcm for one series before and after the update:
\n
Also try running one of the older visualization notebooks: The output of one of the cells where train_localizers.csv is used has changed
\nhttps://www.kaggle.com/code/perseus7/rsna-exploratory-data-analysis-and-visualization/
UPDATE 1:
\nI assumed that the SOPInstanceUID column in train_localizers.csv (version 2025-09-13, after Data Update 2) still points to the old .dcm filenames (before Data Update 2). By using the saved InstanceNumber values corresponding to the old filenames, I was able to match them with the new .dcm filenames for the same InstanceNumber. After that, my training code started working as before. And the problem was solved.
", "rawMarkdown": "Yesterday I ran the training code that worked well on the data before the update, and I saw that on the updated data the exact same code (without any changes) performs terribly.\n\nHas anyone else encountered a similar issue?\n\nI tried to figure out what happened and noticed that the .dcm files seem to have been renamed, but the values in the SOPInstanceUID column of train_localizers.csv have essentially not changed. Seems like the SOPInstanceUID column in train_localizers.csv is pointing to old dcm filenames (before Data Update 2).\n\nBelow you can see my dicts (which I use to sort the .dcm files) InstanceNumber: filename.dcm for one series before and after the update:\n\n\n\n\n\n\nAlso try running one of the older visualization notebooks: The output of one of the cells where train_localizers.csv is used has changed\nhttps://www.kaggle.com/code/perseus7/rsna-exploratory-data-analysis-and-visualization/\n\n\n\n\n\n**UPDATE 1:**\n\nI assumed that the SOPInstanceUID column in train_localizers.csv (version 2025-09-13, after Data Update 2) still points to the old .dcm filenames (before Data Update 2). By using the saved InstanceNumber values corresponding to the old filenames, I was able to match them with the new .dcm filenames for the same InstanceNumber. After that, my training code started working as before. And the problem was solved.", "votes": 3 }, { "id": 3288232, "postDate": "2025-09-13T12:28:42.520Z", "content": "There are a lot of changes in all the data.
", "rawMarkdown": "There are a lot of changes in all the data." } ], "comments": [ { "id": 3288232, "author_name": "ktt proc", "author_url": "", "post_date": "2025-09-13T12:28:42.520000", "content": "There are a lot of changes in all the data.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3288218": "Yesterday I ran the training code that worked well on the data before the update, and I saw that on the updated data the exact same code (without any changes) performs terribly.\n\nHas anyone else encountered a similar issue?\n\nI tried to figure out what happened and noticed that the .dcm files seem to have been renamed, but the values in the SOPInstanceUID column of train_localizers.csv have essentially not changed. Seems like the SOPInstanceUID column in train_localizers.csv is pointing to old dcm filenames (before Data Update 2).\n\nBelow you can see my dicts (which I use to sort the .dcm files) InstanceNumber: filename.dcm for one series before and after the update:\n\n\n\n\n\n\nAlso try running one of the older visualization notebooks: The output of one of the cells where train_localizers.csv is used has changed\nhttps://www.kaggle.com/code/perseus7/rsna-exploratory-data-analysis-and-visualization/\n\n\n\n\n\n**UPDATE 1:**\n\nI assumed that the SOPInstanceUID column in train_localizers.csv (version 2025-09-13, after Data Update 2) still points to the old .dcm filenames (before Data Update 2). By using the saved InstanceNumber values corresponding to the old filenames, I was able to match them with the new .dcm filenames for the same InstanceNumber. After that, my training code started working as before. And the problem was solved.", "3288232": "There are a lot of changes in all the data." } }