Torchvision in NGC Pytorch built without nvJPEG

Accelerated Computing NGC GPU Cloud Docker and NVIDIA Docker
1

In the Pytorch container (nvcr.io/nvidia/pytorch:24.10-py3), when I try to decode a JPEG with nvJPEG torchvision.io.decode_jpeg(), I get the following error:
RuntimeError: decode_jpegs_cuda: torchvision not compiled with nvJPEG support

How would I go about fixing this?
Wouldn’t it make more sense if torchvision was built with nvJPEG by default, since it’s already in the container?

Thanks!

2

Turns out, it’s missing from CMakeLists.txt in the torchvision repo and should probably be fixed there.

3

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.