Current implementation is custom-written. Consider using official torchvision NMS:
Reference: https://github.com/pytorch/vision/blob/main/torchvision/csrc/ops/cpu/nms_kernel.cpp
Benefits:
- Optimized CPU/CUDA implementations available
- Better maintained and tested by PyTorch team
- Reduces custom code maintenance burden
Trade-off: Adds torchvision as dependency (currently only use LibTorch)
Line: 114
|
return y; |
|
} |
|
|
|
// TODO: Replace custom NMS with torchvision library implementation |
|
// category=Vision estimate=4h |
|
// Current implementation is custom-written. Consider using official torchvision NMS: |
|
// Reference: https://github.com/pytorch/vision/blob/main/torchvision/csrc/ops/cpu/nms_kernel.cpp |
|
// Benefits: |
|
// 1. Optimized CPU/CUDA implementations available |
|
// 2. Better maintained and tested by PyTorch team |
|
// 3. Reduces custom code maintenance burden |
Current implementation is custom-written. Consider using official torchvision NMS:
Reference: https://github.com/pytorch/vision/blob/main/torchvision/csrc/ops/cpu/nms_kernel.cpp
Benefits:
Trade-off: Adds torchvision as dependency (currently only use LibTorch)
Line: 114
ReUseX/libs/reusex/src/vision/utils.cpp
Lines 111 to 121 in f6e1667