QuadTok: Quadtree Visual Tokenizer for Autoregressive Image Generation

Hugging Face Daily Papers Papers
autoregressive-generation visual-tokenization quadtree image-generation vision-transformer computer-vision

Summary

QuadTok introduces a hierarchical quadtree-based visual tokenizer that dynamically allocates tokens to intricate image regions, saving ~10% of tokens on ImageNet while enabling a 947M GPT-style model to reach 2.08 gFID on ImageNet 256x256 and zero-shot spatially controlled generation.

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.
Original Article
View Cached Full Text

Cached at: 10/08/26, 12:15 PM

Paper page - QuadTok: Quadtree Visual Tokenizer for Autoregressive Image Generation

Source: https://huggingface.co/papers/2610.10497

Abstract

WeintroduceQuadTok,anovelframeworkforvisualtokenizationandautoregressiveimagegeneration.Comparedtotraditionalapproachesusing2Dgridsor1Dtokensequences,weproposeahierarchicalquadtreestructure,bridgingthegapbetween2Dspatialbindingand1Dsequence-levelflexibility.TheQuadToktokenizerdynamicallyallocatesrepresentationalcapacitytovisuallyintricateareaswhileleavinghomogeneousregionsatacoarseresolution.Comparedwithafixed256-tokengrid,ourImageNet-trainedtokenizersavesapproximately10%oftokensonImageNetand9%whentransferredzero-shottotheCOCOdataset,whilemaintainingcomparablereconstructionfidelity.Furthermore,thenaturalcausalityintroducedbythetreestructureseamlesslyenablesautoregressiveimagegeneration.Conditionedonaquadtreetopologysuppliedbeforegeneration,our947MGPT-stylegenerativemodelachievesa2.08gFIDontheImageNet256times256benchmark.Additionally,leveragingthestrongspatialcorrelationpreservedbythequadtreestructure,theQuadTokgeneratorenableszero-shotspatiallycontrolledimagegenerationcapabilities.Code:https://github.com/myc634/QuadTok.

View arXiv pageView PDFGitHub0Add to collection

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.10497 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.10497 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.10497 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

SemanTok: Predictable Semantic Tokens for Efficient Autoregressive Video Generation

Hugging Face Daily Papers

SemanTok introduces a flexible-length video tokenizer that feeds frozen DINO features into its encoder and reconstructs them from each retained token prefix, enabling predictable coarse-to-fine semantic tokens for autoregressive video generation. A 201M SemanTok AR model matches or beats a VideoFlexTok model 3.4x its size, with strong semantic alignment and fidelity across model scales and noise levels.