timm (PyTorch Image Models) is an open-source library of ready-made computer vision models, most with pretrained weights, plus the tools to train and test them.
In plain words
Building an image model from scratch takes a lot of work and data. timm collects image model families, such as vision transformers and convolutional networks, in one place, with more than 700 pretrained models you can load by name. It also includes the pieces around them: layers, optimizers, learning rate schedules, data loaders, augmentations and training scripts. Its models can serve as a backbone for your own tasks.
A quick example
Imagine you want to sort photos of plants into species. Instead of designing a network, you ask timm for a pretrained model by name, swap its final layer for one with your number of species, and fine-tune it on your own photos.
Where you'll meet it
- Image classification and fine-tuning.
- Backbones for detection and other vision tasks.
Related
Sources
- Docstimm, Hugging Face documentation · read 29 Sept 2026
- Docshuggingface/pytorch-image-models, Hugging Face (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.