What is the relationship between torchvision and torch?

torchvision is a PyTorch package for computer vision tasks, built as an additional library on top of the PyTorch deep learning framework. It offers many tools and functions for processing image and video data. The torchvision library provides various functions required for common computer vision tasks, including data loading, preprocessing, transformation, model definition, training, and evaluation. It also includes popular image and video datasets such as MNIST, CIFAR-10, and ImageNet, along with many pretrained models for tasks like image classification, object detection, and semantic segmentation. Unlike the torch library which focuses on core functionality of the PyTorch framework like tensor computation, automatic differentiation, and optimizers, torchvision is specifically designed for computer vision tasks and utilizes the capabilities of torch to implement its relevant functions and tools. Therefore, torchvision can be seen as an extension of the torch library aimed at supporting computer vision tasks.