Home Original page

bump albumentation to >2 and fix breaking changes by arashsm79 · Pull Request #3174 · DeepLabCut/DeepLabCut

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR updates the albumentations dependency from version <=1.4.3 to >=2 and addresses all breaking API changes. The migration includes parameter name changes, method signature updates, and consistent handling of keypoints/bboxes as numpy arrays throughout the codebase.

Changes:

  • Updated albumentations dependency to >=2 in setup.py and requirements.txt
  • Migrated all deprecated API calls (always_apply → p=1.0, get_params_dependent_on_targets → get_params_dependent_on_data, etc.)
  • Changed keypoint and bbox handling to consistently use numpy arrays instead of lists

Reviewed changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file
File Description
setup.py Updated albumentations constraint from <=1.4.3 to >=2
requirements.txt Updated albumentations constraint from <=1.4.3 to >=2
deeplabcut/pose_estimation_pytorch/data/transforms.py Comprehensive migration of all albumentations API calls to v2 syntax
deeplabcut/pose_estimation_pytorch/data/utils.py Convert keypoints to numpy array before passing to transforms
tests/pose_estimation_pytorch/data/test_transforms.py Updated keypoint format to use lists (converted to arrays internally) and adjusted precision tolerance
tests/pose_estimation_pytorch/other/test_custom_transforms.py Updated keypoint format to numpy array
Comments suppressed due to low confidence (2)

deeplabcut/pose_estimation_pytorch/data/transforms.py:485

  • The ElasticTransform class still passes the deprecated always_apply parameter to the parent A.ElasticTransform class. In albumentations v2, this parameter has been removed and should be converted to p=1.0 if always_apply else p before calling super().init(), similar to other transforms in this file (see Grayscale, CoarseDropout, ScaleToUnitRange examples).
        super().__init__(
            alpha,
            sigma,
            alpha_affine,
            interpolation,
            border_mode,
            value,
            mask_value,
            always_apply,
            approximate,
            same_dxdy,
            p,
        )

deeplabcut/pose_estimation_pytorch/data/transforms.py:437

  • Empty error message in ValueError. Should provide a descriptive message such as 'alpha must be a float, int, or tuple of two numbers.' to help users understand what went wrong.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.