[RFC] Different menaning of angle parameter in affine and rotate ops
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 17.9k
- Forks
- 7.3k
- Avg merge
- 1d 15h
- Merged PRs (30d)
- 13
Description
In Torchvision stable API, there are 2 ops: affine and rotate that has angle as parameter.
However, positive angle parameter in F.affine op performs clockwise rotation.
while in F.rotate it is an opposite meaning, it does counter clockwise rotation:
For Detectron2, Pillow or albumentations, they do counter clockwise rotation for positive angle value.
Question: Should we align F.affine op to perform counter clockwise rotation for positive angle value ?
cc @vfdev-5 @datumbox
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Compare the affine and rotate implementations in torchvision/transforms/functional.py at the linked locations, then review how the angle convention is described for each operation. Check the cited Detectron2, Pillow, and albumentations conventions before proposing an API decision. Done means the convention is agreed and the affected documentation, tests, and behavior changes are identified.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- computer-vision
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 25/100