Incompatible TorchVision Version for PyTorch 2.2.0 on Nvidia Jetson Orin Nano

Issue Overview

Users attempting to install TorchVision compatible with PyTorch v2.2.0 on the Nvidia Jetson Orin Nano development board are encountering difficulties. The specific problem arises when trying to clone the TorchVision repository, as the command to clone branch v0.17 fails with the error message "fatal: Remote branch v0.17 not found in the upstream origin". This issue occurs during the setup process, particularly after successfully installing PyTorch v2.2.0 on the Jetson platform running JetPack 6.0 DP (L4T R36.2.0) with Python 3.10.

Possible Causes

1. Mismatched Versions: The TorchVision version specified (v0.17) may not exist or be compatible with PyTorch v2.2.0.

2. Outdated Installation Instructions: The provided installation steps might be outdated or not applicable to the latest PyTorch version.

3. Repository Changes: The TorchVision repository structure or branch naming convention may have changed since the instructions were written.

4. Network Issues: There could be connectivity problems preventing the successful cloning of the repository.

Troubleshooting Steps, Solutions & Fixes

1. Check Available TorchVision Versions:

   • Visit the official TorchVision repository on GitHub to view available tags and branches.
   • Look for the most recent version that is compatible with PyTorch 2.2.0.

2. Use Release Candidate (RC) Version:

   • As suggested in the forum, try using the latest release candidate version of TorchVision.
   • Modify the clone command to use the appropriate RC tag. For example:

     git clone --branch v0.18.0-rc1 https://github.com/pytorch/vision torchvision
     

3. Update Installation Steps:

   • Replace the following lines in your installation script:

     git clone --branch v0.17 https://github.com/pytorch/vision torchvision
     cd torchvision
     export BUILD_VERSION=0.17.0
     

   • With the updated version information:

     git clone --branch v0.18.0-rc1 https://github.com/pytorch/vision torchvision
     cd torchvision
     export BUILD_VERSION=0.18.0
     

4. Check Network Connection:

   • Ensure you have a stable internet connection.
   • Try using a different network if possible.
   • Verify that you can access GitHub and other online resources.

5. Use Pre-built Wheels:

   • Instead of building from source, check if there are pre-built wheels available for your specific Jetson platform and PyTorch version.
   • Search for compatible wheels on the NVIDIA Developer website or PyPI.

6. Consult Official Documentation:

   • Review the latest installation instructions on the official PyTorch and TorchVision documentation.
   • Check for any specific guidelines for Jetson platforms or ARM64 architectures.

7. Consider Downgrading PyTorch:

   • If no compatible TorchVision version is available for PyTorch 2.2.0, consider downgrading to a stable PyTorch version with known TorchVision compatibility.

8. Seek Community Support:

   • If the issue persists, reach out to the PyTorch community forums or the NVIDIA Jetson developer community for more specific assistance.
   • Provide detailed information about your setup, including exact versions of JetPack, PyTorch, and any error messages encountered.

By following these steps, you should be able to resolve the TorchVision installation issue and successfully set up a compatible version with PyTorch 2.2.0 on your Nvidia Jetson Orin Nano development board.