pytorch/vision

Torchvision is a Python library that provides the building blocks AI developers need when working on computer vision, which means tasks like recognizing objects in photos, classifying images, or detecting faces. It is part of the PyTorch ecosystem, a popular framework for building and training machine learning (AI) models. The library bundles three main things together. First, it gives you access to well-known datasets, public collections of labeled images used to train and test AI models, with tools to automatically download and prepare them. Second, it provides model architectures, which are pre-built AI model designs that have already been trained on large amounts of data so you can start with a capable model rather than building one from scratch. Third, it includes image transformations, functions that preprocess or manipulate images before feeding them into a model, such as resizing, cropping, or adjusting colors. You would use torchvision if you are building a computer vision project in Python and want ready-made components rather than writing everything yourself. It supports image processing via torch tensors (PyTorch's core data format) and PIL images (a common Python image format). The library is written in Python and is maintained by the PyTorch team. Note that while the library helps you download public datasets, the project itself does not host or vouch for those datasets, it is your responsibility to check their licensing before use.