Implement offset regression #447
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
* Fix topdown colorspace preprocessing - Switch to using static shape when cropping. Note that this means the shape of the inputs must be statically known. * Clean up and document nn.inference module - Add most of the missing docstrings and type hints - Remove unused code - Standardized `Predictor` methods - API change: `TopDownModel` -> `TopDownInferenceModel` - API change: `Predictor.predict()` no longer has `make_instances` kwarg
* Base offset map computation * Add support for missing points * Add offset maps masking * Add offset maps to confidence map transformers * Docstring formatting * [offset_regression] Add optional refinement heads to config and training for all models (#451) * Add channel flattening to refinement map generators - Fix output keys * Add offset refinement config options - Disable OHKM by default (unrelated) * Create offset refinement head * Implement model creation with refinement head * Add offset refinement to training pipeline builders - Nit: Re-order classes * Add offset refinement support to trainer - Add minimal tests for trainer (finally) - Rename nn/data/test_training.py -> test_data_training.py * [offset_regression] Add offset refinement to inference (#452) * Add stride arg to offset generation - Needed to scale the raw offsets to sampling grid units * Add global peak finding with learned offsets * Move Hungarian matching utilities to nn.utils * Add local peak finding with learned offsets * Fix offset refinement head from centroid config - Nit: Reorder classes in nn.heads * Fix Model instantiation with offset refinement head from centroid config * Add tests for training centroid models * Add offset refinement to inference module - Nit: Reorder classes - Add attributes to specify index of output tensors for each model - Add detection of head-specific output tensors * Fix typo in bottom-up peak finding * Add minimal trained models as fixtures for tests * Add tests for model loading and inference
Codecov Report
@@ Coverage Diff @@
## develop #447 +/- ##
===========================================
+ Coverage 60.52% 64.14% +3.62%
===========================================
Files 120 121 +1
Lines 18008 18250 +242
===========================================
+ Hits 10899 11707 +808
+ Misses 7109 6543 -566
Continue to review full report at Codecov.
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This feature enables peak location refinement for more accurate learnable subpixel localization.