c++ libtorch jit load failed
dhkdnduq opened this issue · 1 comments
dhkdnduq commented
Describe the bug
boxes.py
41 line
_assert_has_ops() <-- c++ assert
return torch.ops.torchvision.nms(boxes, scores, iou_threshold)
if use soft_nms ==true
libtorch load ok. but so slow
how can i solve this problem?
Desktop (please complete the following information):
- OS: [e.g. Windows]
- PyTorch version w/ CUDA/cuDNN [e.g. from
conda list
, 1.7.0 py3.8_cuda11.0.221_cudnn8.0.3_0] - Pytorch & libtorch v 1.7.1 , py3.8_cuda11.2_cudnn8
Additional context
solved this problem!
using c++ torchvision
https://github.com/pytorch/vision/tree/master/torchvision
also hopes that the onnx problem will be solved soon
rwightman commented
Not an issue here, I think it's an issue with libtorch/torchvision ops for NMS. Possibly fixed in latest torchvision as per SO link below.