Feat: Add finetuning code & dice loss support for IO#62
Open
basardemir wants to merge 16 commits intomainfrom
Open
Feat: Add finetuning code & dice loss support for IO#62basardemir wants to merge 16 commits intomainfrom
basardemir wants to merge 16 commits intomainfrom
Conversation
- Add config-based multi-dataset finetuning with weighted sampling - Support paired/unpaired datasets with optional segmentations - Implement Dice loss for anatomical structure alignment - Add comprehensive finetuning documentation and examples - Include dataset caching and auto-download of pretrained weights - Add per-dataset validation and TensorBoard logging
added 3 commits
December 18, 2025 08:28
Add numpy<2.0 constraint to prevent AttributeError with np.bool. NumPy 2.0+ removed the deprecated np.bool alias that older ITK versions depend on.
lintian-a
reviewed
Jan 11, 2026
| Training progress is logged to TensorBoard: | ||
|
|
||
| ```bash | ||
| tensorboard --logdir=results/my_finetuning_experiment |
Collaborator
There was a problem hiding this comment.
Are the log files saved in the specified output folder, such as " tensorboard --logdir=my_config[training][output_folder]/my_finetuning_experiment"?
lintian-a
reviewed
Jan 11, 2026
| ```bash | ||
| # Navigate to finetuning directory | ||
| cd uniGradICON/ | ||
| pip install . |
Collaborator
There was a problem hiding this comment.
Do we need to consider users who install using "pip install unigradicon"?
lintian-a
reviewed
Jan 11, 2026
| return 'standard' | ||
|
|
||
|
|
||
| def create_dataset_from_config(dataset_config: Dict[str, Any], input_shape: Tuple[int, ...]) -> dataset.Dataset: |
Collaborator
There was a problem hiding this comment.
Do we need to validate that the image files in the dataset JSON file exist?
lintian-a
reviewed
Jan 11, 2026
| desired_coordinate_orientation=desired_coordinate_orientation, | ||
| use_image_direction=True) | ||
|
|
||
| class Dataset: |
Collaborator
There was a problem hiding this comment.
Is this Dataset class inherited from torch.utils.data.Dataset?
lintian-a
reviewed
Jan 11, 2026
| samples_per_epoch = config['training'].get('samples_per_epoch', total_samples) | ||
|
|
||
| total_weight = sum(weights) | ||
| normalized_weights = [w / total_weight for w in weights] |
Collaborator
There was a problem hiding this comment.
Do we want the weight to balance the sizes of the datasets, namely, by sampling more data from the smaller dataset? The current implementation seems to assign a lower weight to datasets of smaller size.
… dataset file validation, refactor datasets to torch base classes, and correct weighted sampling.
…egmentations in TensorBoard, update docs
* Test build * Prepare test setting * Add 'Add_build_workflow' branch to trigger build and deploy workflow * Add step in the workflow to find an empty gpu * 1. improve the GPU selection logic 2. Run CLI test with CPU * Enhance GPU selection logic with detailed logging and re-evaluation for optimal GPU choice * Use another GPU. * Fix CUDA_VISIBLE_DEVICES environment variable type for GPU unit tests * Remove explicit GPU environment variable for unit tests to allow dynamic selection * Update build workflow to require upload options * Refactor build workflow: rename job, update upload steps, and clean up commented code * Update build workflow: remove upload_to_pypi input and comment out TestPyPI conditions * Fix typo. * fix bug * Revert the version back to 1.0.4 * Fix the version number * Refactor workflow: remove process.exit call and rename step to download test data
…TensorBoard viz, Learn2Reg examples, and fix Dice background/memory leak/DICOM/shape bugs
…ture finetuning to dict-based pipeline with 2 dataset types
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
No description provided.