Issues

• batched_nms adds the per-class offsets in half precision: float16 boxes round to multiples of up to 32 px and overflow to inf, so the wrong boxes are kept

  pytorch/vision#9697 · Filter by author Nicholas022400701 opened on Oct 6, 2026

• box_convert from xyxyxyxy to xywhr returns inf for float16 boxes with a side of 256 pixels or more (the squares of the sides overflow)

  pytorch/vision#9695 · Filter by author Nicholas022400701 opened on Oct 6, 2026

• sanitize_bounding_boxes computes the wrong height for rotated boxes

  pytorch/vision#9689 · Filter by author u7k4rs6 opened on Sep 30, 2026

• clamp_bounding_boxes collapses slightly tilted rotated boxes to a point because _order_bounding_boxes_points does not pick the leftmost corner

  pytorch/vision#9676 · Filter by author Nicholas022400701 opened on Sep 25, 2026

• v2 KeyPoints: flip and resize assume different pixel conventions

  pytorch/vision#9675 · Filter by author MaverickTopG opened on Sep 23, 2026
  1

• [RFC] Fused RoIAlign Face Embedding Extractor & Numerically Stable Angular Margin Loss (torchvision.ops & torchvision.losses)

  pytorch/vision#9670 · Filter by author powerofaisinstudy-debug opened on Sep 14, 2026
  1

• Optional ultrafast COCO evaluation backend in Detection reference scripts

  pytorch/vision#9665 · Filter by author developer0hye opened on Sep 11, 2026

• [Feature Request] add ImageFolder/DatasetFolder: incremental index caching to avoid full filesystem re-scans

  pytorch/vision#9659 · Filter by author powerofaisinstudy-debug opened on Sep 6, 2026
  4

• [Feature Request] Add transform pipeline inspection and visualization utilities

  pytorch/vision#9657 · Filter by author kanaad-lims opened on Sep 5, 2026
  3

• Support PIL images in v2.GaussianNoise

  pytorch/vision#9652 · Filter by author YeonwooSung opened on Sep 5, 2026
  1

• XYXY to CXCYWH conversion overflows for narrow dtypes

  pytorch/vision#9594 · Filter by author aswanth-07 opened on Aug 9, 2026
  3

• ElasticTransform is slow for larger images and/or sigma

  pytorch/vision#9577 · Filter by author Callidior opened on Aug 3, 2026

• transforms.v2.Normalize caused small H2D and cpu waits gpu if process image on gpu

  pytorch/vision#9576 · Filter by author NewUserHa opened on Aug 1, 2026

• torchvision.transforms.v2.functional.pad should have maintained the channels_last format of the input

  pytorch/vision#9560 · Filter by author NewUserHa opened on Jul 18, 2026
  2

• Make to_image reading return tensors in CHW

  pytorch/vision#9547 · Filter by author NewUserHa opened on Jul 9, 2026

• Image decoders feature support

  pytorch/vision#9542 · Filter by author NicolasHug opened on Jul 6, 2026
  2

• datasets: CIFAR10 download slow

  pytorch/vision#9536 · Filter by author sjfleming opened on Jul 1, 2026
  1

• Container for multi-pass and randomized subset transform loops

  pytorch/vision#9532 · Filter by author C-you-know opened on Jun 25, 2026
  6

• Path traversal in TorchVision dataset archive extraction

  pytorch/vision#9517 · Filter by author quart27219 opened on Jun 17, 2026

• viable/strict updating silently broken

  pytorch/vision#9501 · Filter by author benjaminglass1 opened on May 26, 2026
  2

• TestRotatedBoxIou::test_iou[xyxyxyxy-float32] fails on arm64 Linux due to float32 precision in format round-trip

  pytorch/vision#9499 · Filter by author happyaron opened on May 26, 2026
  1

• Problem with mask

  pytorch/vision#9471 · Filter by author kalibovers opened on Apr 9, 2026
  2

• Cannot export detection model to onnx with dynamo option

  pytorch/vision#9462 · Filter by author Rayndell opened on Mar 30, 2026
  2

• Source builds on macOS/Linux missing OpenMP flags — at::parallel_for silently falls back to sequential

  pytorch/vision#9455 · Filter by author developer0hye opened on Mar 27, 2026
  1

• RuntimeError: operator torchvision::nms does not exist

  pytorch/vision#9435 · Filter by author mgiessing opened on Mar 10, 2026
  1