#180180th of 242 in ML frameworks and libraries
scenic
JAX/Flax library for training attention-based vision models at scale
- Stars
- 3.8k
- License
- Apache-2.0
- Last commit
- Oct 2026
- Language
- Python
3,770 on 25 Apr 20263,848 on 14 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Scenic is a JAX and Flax codebase for research on attention-based computer vision models. It provides shared libraries for multi-host, multi-device training (input pipelines, losses, metrics, bipartite matchers, trainers) plus project directories with full training and evaluation loops. It includes reimplemented baselines such as ViT, DETR, CLIP, U-Net and SAM, and code for research projects like ViViT.
Who it is for: Researchers building large-scale vision models in JAX
Strengths
- Includes reproduced baselines: ViT, DETR, Deformable DETR, CLIP, MAE, ResNet, U-Net, SAM
- Input pipelines handle multi-host and multi-device splitting, caching and prefetching
- Projects can be config-only or fork the trainers, losses and models entirely
- Hosts code for published research such as ViViT, TokenLearner and OWL-ViT
Weaknesses
- JAX/Flax only, no PyTorch or TensorFlow support
- Philosophy favors forking and copy-pasting, so code is duplicated across projects
- Many projects need extra packages from their own requirements files
- No tagged releases; install is from a git clone
What it needs
- Needs JAX, Flax, Python 3.9+
- Models: ViT, DETR, CLIP, ResNet, U-Net
Also in ML frameworks and libraries
See all 242| Rank | Project | Score |
|---|---|---|
| 1 | transformersModel-definition library for pretrained text, vision, audio and multimodal models | 94 out of 100 |
| 2 | ultralyticsPython library and CLI for training and running YOLO vision models | 94 out of 100 |
| 3 | pytorch-image-modelsPyTorch library of image models, pretrained weights and training scripts | 94 out of 100 |
| #4 | pytorchPython tensor library with GPU support and tape-based autograd | 93 out of 100 |
| #5 | supervisionPython toolkit for loading, annotating and counting computer vision detections | 92 out of 100 |