more robust ignore label support #543
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
There was a problem hiding this comment.
Pull Request Overview
This PR adds support for configurable ignore label values across the training pipeline. It enables users to specify labels that should be treated as background during patch validation, loss computation, metrics evaluation, and skeleton transform processing.
Key changes:
- Added
zero_ignore_labelsutility function to handle ignore label zeroing in patch validation - Extended loss configuration, metrics initialization, and transforms to accept and use ignore values via aliases (
ignore_index,ignore_label,ignore_value) - Updated dependency versions in
pyproject.toml
Reviewed Changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| vesuvius/src/vesuvius/models/datasets/find_valid_patches.py | Added zero_ignore_labels function and integrated ignore label handling into patch validation logic |
| vesuvius/src/vesuvius/models/datasets/slicers/chunk.py | Updated chunk slicer to propagate ignore values through the validation pipeline |
| vesuvius/src/vesuvius/models/datasets/base_dataset.py | Added _resolve_target_ignore_labels method and integrated ignore values into chunk volumes and skeleton transforms |
| vesuvius/src/vesuvius/models/training/train.py | Enhanced loss initialization and metrics to accept ignore values from target configuration |
| vesuvius/src/vesuvius/models/augmentation/transforms/utils/skeleton_transform.py | Modified skeleton transform to handle ignore values during processing |
| vesuvius/pyproject.toml | Updated package dependencies including version changes and new packages |
Comments suppressed due to low confidence (1)
vesuvius/pyproject.toml:64
- The version
s3fs>=2025.9.0appears to use a future version number. As of January 2025, s3fs follows calendar versioning similar to fsspec. Version 2025.9.0 may not exist yet. Please verify this is the correct version.
"s3fs>=2025.9.0",
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| MedialSurfaceTransform( | ||
| do_tube=False, | ||
| target_keys=skeleton_targets, | ||
| ignore_values=ignore_values or None, |
There was a problem hiding this comment.
The expression ignore_values or None is redundant. An empty dict {} is falsy, so this will return None when ignore_values is empty. Consider simplifying to just ignore_values and let the transform handle the empty dict, or use ignore_values if ignore_values else None to be more explicit about the intent.
| MedialSurfaceTransform( | ||
| do_tube=False, | ||
| target_keys=skeleton_targets, | ||
| ignore_values=ignore_values or None, |
There was a problem hiding this comment.
The expression ignore_values or None is redundant. An empty dict {} is falsy, so this will return None when ignore_values is empty. Consider simplifying to just ignore_values and let the transform handle the empty dict, or use ignore_values if ignore_values else None to be more explicit about the intent.
| ignore_values=ignore_values or None, | |
| ignore_values=ignore_values if ignore_values else None, |
| else: | ||
| try: | ||
| mask = (arr == ignore_label) | ||
| except TypeError: |
There was a problem hiding this comment.
This statement is unreachable.
|
@bruniss what's this? |
No description provided.