PyTorch and torchvision versions are incompatible

[Heartful-echo]

Hi

I am using the below JETPACK with a JETSON AGX Xavier.
JetPack 5.0.2 (L4T R35.1.0)

I tried the NVIDIA L4T PyTorch from the NGC catalog, but the 32GB Xavier didn’t have enough disk space! !
Therefore the container cannot be used.

I installed the whl of PyTorch v1.11.0 by following the forum content below.
https://forums.developer.nvidia.com/t/pytorch-for-jetson/72048

pytorch can use cuda. So far so good.

I installed pytorch v1.11.0, so torchvision installed 0.12.0, see below.
https://github.com/pytorch/vision#installation

I installed torchvision using pip.

However!
When using YOLOv5 it outputs the error below.

RuntimeError: Couldn't load custom C++ ops. This can happen if your PyTorch and torchvision versions are incompatible, or if you had errors while compiling torchvision from source. For further information on the compatible versions, check https://github.com/pytorch/vision#installation for the compatibility matrix. Please check your PyTorch version with torch.__version__ and your torchvision version with torchvision.__version__ and verify if they are compatible, and if not please reinstall torchvision so that it matches your PyTorch install.

It’s very troubled.
help me.

[AastaLLL]

Hi,

Please try our PyTorch1.13.0 nv22.09 with TorchVision 0.15.0.
Details can be found in the below topic:

TensorRT INT8 calibration python API

Hello @AastaLLL , I think I have resolved the problem. Today 05/10/2022 Nvidia has uploaded a new version of Torch+CUDA support compatible with Jetpack 5.0.2. So I have installed the last one and I have build Torchvision from source here. After doing that, I have Torch and TorchVision both with CUDA support I think. [image] I tried and ran the val.py scipt from yolov5 and it worked. Now I wil try to run it using a TensorRT engine, I hope there will be no issues. I will keep you updated. T…

Thanks.

[Heartful-echo]

@AastaLLL

Thanks for your comment.
Where can I get PyTorch1.13.0 nv22.09 and TorchVision 0.15.0?
I want PyTorch and TorchVision in whl format.

Thank you

[Heartful-echo]

I have resolved this issue!!

I’ll record the command I typed for reference.
I use PyTorch1.11.0 so torchvision specified 0.12.0. This is the rule.

git clone --recursive --branch v0.12.0 https://github.com/pytorch/vision --depth 1
cd vision
python3 setup.py install --user
pip3 uninstall torch torchvision
sudo pip3 install torch-1.11.0-cp38-cp38-linux_aarch64.whl

pip3 list

I got YOLOv5 working. It’s working as expected.

[AastaLLL]

Thanks for the feedback.
Good to know it works now.