NicolasHug
Follow
Pinned
915 contributions in the last year
Activity overview
Contribution activity
February 2023
Created 34 commits in 3 repositories
Created 1 repository
- NicolasHug/mobile-vision Python
Created a pull request in pytorch/vision that received 3 comments
Remove usage of torch._six
Follow-up to its removal in torch core pytorch/pytorch#94709
+2
−2
•
3
comments
Opened 25 other pull requests in 2 repositories
pytorch/vision
20
merged
1
open
3
closed
- Remove check_v2_dataset_warnings.py which is failing
-
Remove old TODO about using
_log_api_usage_once() - Make rotate_image_tensor's results fully BC with v1
- Add a removable warning in torchvision.transforms.v2 and torchvision.datapoints
- Deprecate functional_pil and functional_tensor and make them private
- Make RandomIoUCrop compatible with SanitizeBoundingBoxes
- Copy paste prototype.datapoints and prototype.transforms out of prototype area
- Make RandomApply torchscriptable in V2
- Add SanitizeBoundingBoxes transform
- Remove unused import
- Let LinearTransformation return datapoints instead of tensors
-
Use
is Falsefor some antialias checks - Tighter test against PIL on resize
- Add tests for transform presets, and various fixes
- Prototype references
- Skip model test for vit_h_14
- Try GPU tests on 11.6 and 11.8
- Use larger runner and only test vit_h_14
- Increase tolerance fo resnet101 model test
- Test some flaky detection models on float64 instead of float32
- Put back previous tolerance for test_classification and test_video
- Fix resnet_fpn_backbone(pretrained=True)
- Change default of antialias parameter from None to 'warn'
- NOMRG [xplat][torchvision] Add torchvision streaming decoder
facebookresearch/mobile-vision
1
closed
Reviewed 64 pull requests in 2 repositories
pytorch/vision
25 pull requests
- Added wrap_dataset_for_transforms_v2 into datasets and handled beta w…
- Split tests for transforms v2 and prototype
- Add more tests for bounding boxes
- convert all bounding boxes to XYXY in dataset wrapper
- Remove nit comment
- [NOMERGE] drop in transforms v2 into the v1 tests
- Fix dropout issue in swin transformers
- Make RandomIoUCrop compatible with SanitizeBoundingBoxes
- reduce GHA log output
- make type alias private
- Promote prototype transforms to beta status
- fix ten_crop datapoint return
- make fill defaultdict an implementation detail
- Fixed issues in elastic transform
- Make RandomApply torchscriptable in V2
- allow integer parameters in ColorJitter
- Ten crop annotation
- add error for max_size with size sequence in resize
- Add SanitizeBoundingBoxes transform
- Support integer values for interpolation in the prototype transforms
- Fix some annotations in transforms v2 for JIT v1 compatibility
- add proper smoke test for prototype transforms
- [PoC] add support for instance checks on dataset wrappers
- Undeprecate int constants for interpolation
- allow subclasses in dataset wrappers
- Some pull request reviews not shown.
pytorch/pytorch
1 pull request
Created an issue in pytorch/vision that received 3 comments
Remove usage of scipy.stats.binom_test
It's deprecated. Not super urgent at we only rely on it in the tests.
/work/test/test_transforms.py:1437: DeprecationWarning: 'binom_test' is depr…
3
comments