vision

TorchVision is a computer vision library for the PyTorch ecosystem. It bundles popular public datasets, common image transformations, and neural network model architectures intended for computer vision workflows. The package supports torch tensors and PIL images, including Pillow and the Pillow-SIMD drop-in replacement. Stable installation is handled through the official PyTorch instructions, while source builds are documented in the project contributing guide. The README maintains a compatibility matrix showing matching PyTorch, TorchVision, and Python versions; the current stable line listed is PyTorch 2.13 with TorchVision 0.28 for Python 3.10 through 3.14. TorchVision downloads and prepares public datasets but does not host or distribute them, and users are responsible for checking dataset licenses. Pretrained models may also have separate license terms derived from their training data; for example, the README notes that SWAG models use CC-BY-NC 4.0. API documentation is published on the official PyTorch website, and the repository provides contribution guidance and a citation entry.