license: apache-2.0
pipeline_tag: feature-extraction
UniTok: A Unified Tokenizer for Visual Generation and Understanding
This repository contains UniTok, a unified visual tokenizer for both image generation and understanding tasks, as presented in UniTok: A Unified Tokenizer for Visual Generation and Understanding.
Project Page: https://foundationvision.github.io/UniTok/
Code: https://github.com/FoundationVision/UniTok
UniTok encodes fine-grained details for generation and captures high-level semantics for understanding. It's compatible with autoregressive generative models (e.g., LlamaGen), multimodal understanding models (e.g., LLaVA), and unified MLLMs (e.g., Chameleon and Liquid).
Built upon UniTok, we construct an MLLM capable of both multimodal generation and understanding, which sets a new state-of-the-art among unified autoregressive MLLMs. The weights of our MLLM will be released soon.
Performance
Method | #Tokens | rFID ↓ | Accuracy |
---|---|---|---|
VQVAE Model | |||
VQ-GAN | 256 | 4.98 | -- |
RQ-VAE | 256 | 1.30 | -- |
VAR | 680 | 0.90 | -- |
CLIP Model | |||
CLIP | 256 | -- | 76.2 |
SigLIP | 256 | -- | 80.5 |
ViTamin | 256 | -- | 81.2 |
Unified Model | |||
TokenFlow † | 680 | 1.37 | -- |
VILA-U † | 256 | 1.80 | 73.3 |
UniTok | 256 | 0.39 | 70.5 |
UniTok † | 256 | 0.38 | 78.6 |
† indicates the model uses pretrained CLIP weights for initialization. Although CLIP weight initialization boosts ImageNet zero-shot accuracy, we notice that random initialization leads to better downstream understanding performance. We thus release the model checkpoint of UniTok that is trained from scratch.
Model Weights
Model | Res. | #Token | Code Shape | rFID | Checkpoint |
---|---|---|---|---|---|
UniTok-Large | 256 | 256 | 16 $\times$ 16 $\times$ 8 | 0.39 | Download |
Usage
(... rest of README content ...)
Citation
@article{unitok,
title={UniTok: A Unified Tokenizer for Visual Generation and Understanding},
author={Ma, Chuofan and Jiang, Yi and Wu, Junfeng and Yang, Jihan and Yu, Xin and Yuan, Zehuan and Peng, Bingyue and Qi, Xiaojuan},
journal={arXiv preprint arXiv:2502.20321},
year={2025}
}