| title | QuadTok: Quadtree Visual Tokenizer for Autoregressive Image Generation |
|---|---|
| authors | Yucheng Mao, Zeyuan Chen, Xiaojun Shan, Xiang Zhang, Divyansh Srivastava, Bingnan Li, Zhuowen Tu |
| date | 2026-10-07 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2610.10497 |
| upvotes | 17 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| abstract | We introduce QuadTok, a novel framework for visual tokenization and autoregressive image generation. Compared to traditional approaches using 2D grids or 1D token sequences, we propose a hierarchical quadtree structure, bridging the gap between 2D spatial binding and 1D sequence-level flexibility. The QuadTok tokenizer dynamically allocates representational capacity to visually intricate areas while leaving homogeneous regions at a coarse resolution. Compared with a fixed 256-token grid, our ImageNet-trained tokenizer saves approximately 10% of tokens on ImageNet and 9% when transferred zero-shot to the COCO dataset, while maintaining comparable reconstruction fidelity. Furthermore, the natural causality introduced by the tree structure seamlessly enables autoregressive image generation. Conditioned on a quadtree topology supplied before generation, our 947M GPT-style generative model achieves a 2.08 gFID on the ImageNet 256 times 256 benchmark. Additionally, leveraging the strong spatial correlation preserved by the quadtree structure, the QuadTok generator enables zero-shot spatially controlled image generation capabilities. Code: https://github.com/myc634/QuadTok. |
cite
@misc{mao2026quadtok,
title={QuadTok: Quadtree Visual Tokenizer for Autoregressive Image Generation},
author={Yucheng Mao and Zeyuan Chen and Xiaojun Shan and Xiang Zhang and Divyansh Srivastava and Bingnan Li and Zhuowen Tu},
year={2026},
eprint={2610.10497},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2610.10497}
}