![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2Fbc8b869a7affec2cd291b1f0feb83f05%2FSelection_1498.png?generation=1765048587811199&alt=media\")
The prediction (which may be correct) may go out of the unlabelled region. At evaluation, the unlabelled region is masked and set to zero, leaving a few noisy voxels.
\nThis is considered as FP in Betti numbers.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2F94366280e29d4b72ba37a3762849e193%2FSelection_1499.png?generation=1765049043533832&alt=media\")
The prediction (which may be correct) may go out of the unlabelled region. At evaluation, the unlabelled region is masked and set to zero, leaving a few noisy voxels.
\nThis is considered as FP in Betti numbers.
\n
Thanks for noticing this edge case! Our team of annotators tried to ensure the presence of some real background next to the ignore label, but this scenario could indeed happen. One way to mitigate it could be outputting thinner predictions, so that they will less likely bleed out of a possible ignore region.
", "rawMarkdown": "Thanks for noticing this edge case! Our team of annotators tried to ensure the presence of some real background next to the ignore label, but this scenario could indeed happen. One way to mitigate it could be outputting thinner predictions, so that they will less likely bleed out of a possible ignore region.", "votes": null }, { "id": "3366868", "postDate": "12/08/2025 03:59:10", "content": "Thanks for the suggestion. i will try to make a \"thinner prediction.\"
\nAnother possible solution (but more complicated to implement)
\n@giorgioangelotti \nCurrently, I'm evaluating my model locally. I like to ensure one thing, if the model gives false positive preidction (predicting on invalid region), will it penalize in the score? Or, the prediction must happen in the valid region with proper topology connectivity?
", "rawMarkdown": "giorgioangelotti \nCurrently, I'm evaluating my model locally. I like to ensure one thing, if the model gives false positive preidction (predicting on invalid region), will it penalize in the score? Or, the prediction must happen in the valid region with proper topology connectivity?", "votes": null }, { "id": "3381696", "postDate": "12/25/2025 09:12:22", "content": "I tried to replicate the scoring metrics in https://www.kaggle.com/code/jirkaborovec/replicate-lb-score-topology-aware-3d-surface-segm and it would be great if organizers could have look…
", "rawMarkdown": "I tried to replicate the scoring metrics in https://www.kaggle.com/code/jirkaborovec/replicate-lb-score-topology-aware-3d-surface-segm and it would be great if organizers could have look...", "votes": null } ], "comments": [ { "id": 3365715, "author_name": "giorgioangelotti", "author_url": "", "post_date": "12/07/2025 10:14:34", "content": "Thanks for noticing this edge case! Our team of annotators tried to ensure the presence of some real background next to the ignore label, but this scenario could indeed happen. One way to mitigate it could be outputting thinner predictions, so that they will less likely bleed out of a possible ignore region.
", "votes": null, "replies": [ { "id": 3366868, "author_name": "hengck23", "author_url": "", "post_date": "12/08/2025 03:59:10", "content": "Thanks for the suggestion. i will try to make a \"thinner prediction.\"
\nAnother possible solution (but more complicated to implement)
\n@giorgioangelotti \nCurrently, I'm evaluating my model locally. I like to ensure one thing, if the model gives false positive preidction (predicting on invalid region), will it penalize in the score? Or, the prediction must happen in the valid region with proper topology connectivity?
", "votes": null, "replies": [] }, { "id": 3381696, "author_name": "jirkaborovec", "author_url": "", "post_date": "12/25/2025 09:12:22", "content": "I tried to replicate the scoring metrics in https://www.kaggle.com/code/jirkaborovec/replicate-lb-score-topology-aware-3d-surface-segm and it would be great if organizers could have look…
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3364904": "\n\nThe prediction (which may be correct) may go out of the unlabelled region. At evaluation, the unlabelled region is masked and set to zero, leaving a few noisy voxels.\n\nThis is considered as FP in Betti numbers. \n\n\n", "3365715": "Thanks for noticing this edge case! Our team of annotators tried to ensure the presence of some real background next to the ignore label, but this scenario could indeed happen. One way to mitigate it could be outputting thinner predictions, so that they will less likely bleed out of a possible ignore region.", "3366868": "Thanks for the suggestion. i will try to make a \"thinner prediction.\"\n\nAnother possible solution (but more complicated to implement)\n- first treat unlabelled as a separate connected component in evaluation (e.g. do not mask it out) \n- any predicted component that intersects the unlabelled component >0.95 is removed\n- then remove unlabelled component \n- proceed to compute Betti numbers as usual", "3381385": "giorgioangelotti \nCurrently, I'm evaluating my model locally. I like to ensure one thing, if the model gives false positive preidction (predicting on invalid region), will it penalize in the score? Or, the prediction must happen in the valid region with proper topology connectivity?", "3381696": "I tried to replicate the scoring metrics in https://www.kaggle.com/code/jirkaborovec/replicate-lb-score-topology-aware-3d-surface-segm and it would be great if organizers could have look..." }, "source": "meta" }