added dinov2 #3186
added dinov2 #3186
Conversation
gisilvs
requested review from
SkafteNicki,
Borda and
justusschock
as code owners
for more information, see https://pre-commit.ci
SkafteNicki
added
enhancement
|
I manage to refactor the new dinov2 feature extractor into the existing fid implementation keeping backwards compatibility |
Codecov Report❌ Patch coverage is ❌ Your project check has failed because the head coverage (37%) is below the target coverage (95%). You can increase the head coverage or adjust the target coverage.
Additional details and impacted files@@ Coverage Diff @@
## master #3186 +/- ##
=========================================
- Coverage 99% 37% -62%
=========================================
Files 15 349 +334
Lines 195 19938 +19743
=========================================
+ Hits 194 7376 +7182
- Misses 1 12562 +12561 🚀 New features to boost your workflow:
|
There was a problem hiding this comment.
Pull Request Overview
This PR adds support for DinoV2 feature extractors to the FrechetInceptionDistance metric, expanding beyond the original InceptionV3-only implementation. The changes introduce string-based feature specification and deprecate integer inputs while maintaining backward compatibility.
- Adds
NoTrainDinoV2class for DinoV2 feature extraction - Introduces string-based feature specification (e.g., "inception-2048", "dino-768")
- Updates API to use
feature_extractorinstead ofinceptionattribute names
Reviewed Changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| src/torchmetrics/image/fid.py | Core implementation adding DinoV2 support, feature mapping, and API updates |
| tests/unittests/image/test_fid.py | Comprehensive test updates for new feature extractors and string-based configurations |
| CHANGELOG.md | Documents the new DinoV2 feature extractor support |
Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.
| self.inception = NoTrainInceptionV3( | ||
| name="inception-v3-compat", | ||
| self.feature_extractor = NoTrainInceptionV3( | ||
| name=f"inception-{feature}", |
There was a problem hiding this comment.
The name parameter should be 'inception-v3-compat' to maintain compatibility with the original implementation, not f'inception-{feature}'. This change could break existing functionality that depends on the specific model name.
| name=f"inception-{feature}", | |
| name="inception-v3-compat", |
Copilot uses AI. Check for mistakes.
| ], | ||
| ) | ||
| def test_compare_fid(tmpdir, equal_size, feature_config): | ||
| """Check that the whole pipeline gives the same result as torch-fidelity.""" |
There was a problem hiding this comment.
Typo in comment: 'whole' should be 'hole' to maintain consistency with the original comment.
Copilot uses AI. Check for mistakes.
What does this PR do?
Fixes #3110
Before submitting
PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
Did you have fun?
Make sure you had fun coding 🙃
📚 Documentation preview 📚: https://torchmetrics--3186.org.readthedocs.build/en/3186/