How to install Torch and Torchvision with cuda supported in Jetpack 5.1.2

Autonomous Machines Jetson & Embedded Systems Jetson Xavier NX
1

I tried many processes to install a compatible torchvision version for CUDA-supported Torch on my Jetson DSBOX-NX2 (Jetpack 5.1.2), but I failed every time. I followed the instructions here: Installing PyTorch for Jetson Platform - NVIDIA Docs

where I found the process to install CUDA-supported Torch. However, I couldn’t install a compatible Torch version.

Then, I tried using a Docker container from this link:
https://github.com/dusty-nv/jetson-containers/tree/master/packages/pytorch.
I used ./autotag, a specific version, docker run, and manually built every process, but in every container, I only got Torch installed, and there was no torchvision installed.

Please provide me with a process to install Torch and torchvision on my Jetson DSBOX-NX2 (Jetpack 5.1.2) so that I can easily use Torch and torchvision with this device.

2

@jonyghosh444 what are the errors that you get when you try building torchvision like under the Installation section of this thread?

That container only has pytorch - instead use the torchvision container (which includes pytorch + torchvision) or l4t-pytorch (which includes pytorch/torchvision/torchaudio/ect)

4

Are there any instructions for installing compatible torch and torchvision, or the full PyTorch package (including torch, torchvision, and torchaudio) for JetPack 5.1.2 on my base environment, instead of using a Docker container?

5

Hi @jonyghosh444, yes in this thread that I linked to above:

7

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