CUDA&torchvision can't use GPU for object detection and tracking on JETSON orin nano 8GB

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
, ,
1

Hi everyone,
my problem is this:
i can’t use indentification and object tracking with the gpu.
1- when i have all the latest compatible versions i can get my interface to work but cuda can’t use the gpu. when i activate yolo my interface becomes extremely slow because it uses the CPU.
2-When I downgrade the dependencies. CUDA works but I get an error:
TorchVisionError: operator torchvision::nms does not exist
and I can’t launch my interface at all.

I’ve tried many different versions and combinations but nothing works…
either the versions are not compatible
or I can’t launch my interface because of a lack of dependencies
or I can’t use the GPU

does anyone have the right combination of compatible and functional versions?
for PyTorch - TorchVision - cuDNN - CUDA - Python - ultralytics

3

Hi,

You can find a step-by-step command in the below link:
https://elinux.org/Jetson/L4T/TRT_Customized_Example#YoloV11_using_Pytorch

Thanks.

1 Like