Factorized Visual Tokenization and Generation

TL;DR

This paper introduces Factorized Quantization (FQ), a scalable and efficient method for visual tokenization that decomposes large codebooks into sub-codebooks with regularization and leverages pretrained models to enhance image generation.

Contribution

It proposes a novel factorization approach for VQ-based tokenizers, improving scalability, diversity, and semantic richness in visual representations.

Findings

Achieves state-of-the-art reconstruction quality.

Enhances auto-regressive image generation performance.

Reduces codebook complexity and redundancy.

Abstract

Visual tokenizers are fundamental to image generation. They convert visual data into discrete tokens, enabling transformer-based models to excel at image generation. Despite their success, VQ-based tokenizers like VQGAN face significant limitations due to constrained vocabulary sizes. Simply expanding the codebook often leads to training instability and diminishing performance gains, making scalability a critical challenge. In this work, we introduce Factorized Quantization (FQ), a novel approach that revitalizes VQ-based tokenizers by decomposing a large codebook into multiple independent sub-codebooks. This factorization reduces the lookup complexity of large codebooks, enabling more efficient and scalable visual tokenization. To ensure each sub-codebook captures distinct and complementary information, we propose a disentanglement regularization that explicitly reduces redundancy, promoting diversity across the sub-codebooks. Furthermore, we integrate representation learning into the training process, leveraging pretrained vision models like CLIP and DINO to infuse semantic richness into the learned representations. This design ensures our tokenizer captures diverse semantic levels, leading to more expressive and disentangled representations. Experiments show that the proposed FQGAN model substantially improves the reconstruction quality of visual tokenizers, achieving state-of-the-art performance. We further demonstrate that this tokenizer can be effectively adapted into auto-regressive image generation. https://showlab.github.io/FQGAN