pytorch/vision

Datasets, Transforms and Models specific to Computer Vision

What it solves

Torchvision simplifies the development of computer vision applications by providing a centralized collection of essential tools, including pre-built datasets, standard model architectures, and common image processing functions.

How it works

The library integrates directly with the PyTorch ecosystem. It provides utilities to download and prepare public datasets, offers ready-to-use model architectures, and includes transformation tools to process images using backends like PIL or standard torch tensors.

Who it’s for

It is designed for machine learning researchers and developers working with PyTorch who need a reliable, standardized toolkit to accelerate the creation and training of computer vision models.

Highlights

  • Includes a wide range of popular, ready-to-use computer vision datasets.
  • Provides implementations of standard model architectures.
  • Offers flexible image transformation tools for data preprocessing.
  • Supports multiple image backends, including high-performance options like Pillow-SIMD.

Related

  • Project
    facebookresearch/pytorch3d

    A PyTorch-based library for 3D Computer Vision research providing differentiable rendering and efficient tools for manipulating 3D meshes and point clouds.

  • Project
    TorchIO-project/torchio

    A Python library for PyTorch that provides tools to efficiently read, preprocess, augment, and sample 3D medical images for deep learning.

  • Project
    frgfm/holocron

    A PyTorch-based library providing implementations of modern deep learning architectures, loss functions, and optimization techniques for computer vision tasks.

  • Project
    azavea/raster-vision

    A Python library and low-code framework for building computer vision models on satellite, aerial, and drone imagery using PyTorch.

  • Project
    facebookresearch/multimodal

    A PyTorch library for training state-of-the-art multimodal models at scale, providing modular building blocks and pretrained weights for models like CLIP and BLIP-2.