vision · Issues· 1222 open
Open on GitHubLocally synced open issues (discussions stay on GitHub)
- #9657
[Feature Request] Add transform pipeline inspection and visualization utilities
Updated Sep 17, 2026 - #9659
[Feature Request] add ImageFolder/DatasetFolder: incremental index caching to avoid full filesystem re-scans
Updated Sep 16, 2026 - #9670
[RFC] Fused RoIAlign Face Embedding Extractor & Numerically Stable Angular Margin Loss (`torchvision.ops` & `torchvision.losses`)
Updated Sep 14, 2026 - #9665
Optional ultrafast COCO evaluation backend in Detection reference scripts
Updated Sep 11, 2026 - #8188
pil_to_tensor() doesn't work for PIL Image with I;16 mode
bugUpdated Sep 9, 2026 - #9594
XYXY to CXCYWH conversion overflows for narrow dtypes
Updated Sep 8, 2026 - #9029
Rotation sampling grid is undesireably low-res
Updated Sep 5, 2026 - #9652
Support PIL images in v2.GaussianNoise
Updated Sep 5, 2026 - #9387
Investigate merging box conversion utils from ops and transforms
Updated Sep 5, 2026 - #9162
CocoDetection wrapped with wrap_dataset_for_transforms_v2 returns image_id even if not in target_keys
Updated Sep 5, 2026 - #9337
Make "RuntimeError: operator torchvision::nms does not exist" error better
Updated Sep 5, 2026 - #9151
Follow-up on Pillow's mode argument in Image.fromarray()
Updated Sep 5, 2026 - #7349
test_schema_meta_validation[maskrcnn_resnet50_fpn_v2] needs to be fixed
Updated Sep 5, 2026 - #8141
`to_pil_image` rounds down
Updated Sep 5, 2026 - #9050
Add Unlabeled Image Dataset for Unsupervised Training
Updated Aug 29, 2026