Skip to main content

Torchvision

Torchvision is a PyTorch companion library that provides popular datasets, model architectures, and image transformations for computer vision tasks.

This document explains how to install and verify Torchvision on NVIDIA Jetson platforms.

1. System Information​

Before installation, confirm your system environment.

System Information

2. Install Dependencies​

Update the system:

sudo apt update

Install required build dependencies:

sudo apt install -y ninja-build libwebp-dev libjpeg-dev

3. Install Torchvision​

Torchvision versions must match the installed PyTorch version.

Version Compatibility​

torch torchvision Python

main / nightly main / nightly >=3.9, <=3.12 2.5 0.20 >=3.9, <=3.12

For PyTorch 2.5, Torchvision 0.20 is required.

3.1 Offline Installation​

Download the Torchvision wheel file manually:

Download URL (external):
https://github.com/ultralytics/assets/releases/download/v0.0.0/torchvision-0.20.0a0+afc54f7-cp310-cp310-linux_aarch64.whl

Navigate to the download directory:

cd ~/Downloads

Install Torchvision:

sudo pip3 install torchvision-0.20.0a0+afc54f7-cp310-cp310-linux_aarch64.whl

3.2 Online Installation​

Install Torchvision directly via pip:

sudo pip3 install https://github.com/ultralytics/assets/releases/download/v0.0.0/torchvision-0.20.0a0+afc54f7-cp310-cp310-linux_aarch64.whl

3.3 Source Code Compilation (Advanced)​

Download the Torchvision source code (v0.20):

https://github.com/pytorch/vision/tree/v0.20.0

Enter the extracted directory and build from source:

cd vision-0.20.0
sudo python3 setup.py install
note

Source compilation is recommended only if prebuilt wheels are unavailable or customization is required.

4. Verify Installation​

Verify that Torchvision is installed correctly:

python3 -c "import torchvision; print(f'Torchvision: {torchvision.__version__}')"

Expected output:

Torchvision Version Output

References​

Maintained by HemiHex for Jetson-based advanced vision workflows.