Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Image Generation

TL;DR

This paper introduces VFMTok, a novel image tokenizer built on a frozen vision foundation model, improving image generation quality, efficiency, and convergence speed through region-adaptive quantization and semantic alignment.

Contribution

It presents a new approach to building image tokenizers on frozen vision models, with innovative quantization and semantic reconstruction techniques.

Findings

Achieves a gFID of 1.36 on ImageNet benchmarks.

Accelerates autoregressive model convergence by three times.

Enables high-fidelity class-conditional image synthesis without classifier-free guidance.

Abstract

In this work, we present a novel direction to build an image tokenizer directly on top of a frozen vision foundation model, which is a largely underexplored area. Specifically, we employ a frozen vision foundation model as the encoder of our tokenizer. To enhance its effectiveness, we introduce two key components: (1) a region-adaptive quantization framework that reduces redundancy in the pre-trained features on regular 2D grids, and (2) a semantic reconstruction objective that aligns the tokenizer's outputs with the foundation model's representations to preserve semantic fidelity. Based on these designs, our proposed image tokenizer, VFMTok, achieves substantial improvements in image reconstruction and generation quality, while also enhancing token efficiency. It further boosts autoregressive (AR) generation -- achieving a gFID of 1.36 on ImageNet benchmarks, while accelerating model convergence by three times, and enabling high-fidelity class-conditional synthesis without the need for classifier-free guidance (CFG). The code is available at https://github.com/CVMI-Lab/VFMTok.