Rotated bounding box NMS implementation for GPU#9478

Draft

zy1git wants to merge 26 commits into

pytorch:mainfrom

zy1git:rotated-NMS-GPU

Commits on Mar 23, 2026

NMS implementation for CPU
Zhitao Yu
committed
add tests
Zhitao Yu
committed
Merge remote-tracking branch 'origin/main' into rotated-NMS
Zhitao Yu
committed

Commits on Mar 24, 2026

fuse two implementations
Zhitao Yu
committed
Remove Detectron2 license header from nms_kernel.cpp since the NMS algorithm is standard TorchVision code; attribution already in box_iou_rotated_utils.h
Zhitao Yu
committed
revert the unnecessary linting
Zhitao Yu
committed
Preserve original IoU computation in fused NMS implementation
Zhitao Yu
committed
Add box caching in fused NMS to match original memory access pattern
Zhitao Yu
committed
fix the test failure
Zhitao Yu
committed

Commits on Mar 25, 2026

remove the torchscript test and err_msg
Zhitao Yu
committed
Remove the fmt parameter from nms function
Zhitao Yu
committed
fix the test failures
Zhitao Yu
committed

Commits on Mar 27, 2026

Reuse TestNMS._reference_nms in TestNMSRotated
Zhitao Yu
committed
address the comment
Zhitao Yu
committed
change the variable name
Zhitao Yu
committed
address more comments on the file torchvision/csrc/ops/cpu/nms_kernel.cpp
Zhitao Yu
committed
add batched_nms and the test
Zhitao Yu
committed

Commits on Mar 31, 2026

Rename _reference_nms to _reference_aligned_nms
Zhitao Yu
committed
Merge TestNMSRotated into TestNMS
Zhitao Yu
committed

Commits on Apr 1, 2026

Commits on Apr 3, 2026

Commits on Apr 16, 2026

Rotated bounding box NMS implementation for GPU
Zhitao Yu
committed