Channel-wise Vector Quantization

2026-05-25

AI Summary

This work addresses the limitations of conventional image tokenization methods, which rely on spatial patch-based vector quantization and struggle to capture hierarchical visual structures effectively. The authors propose Channel-wise Vector Quantization (CVQ), a novel approach that shifts discretization from the spatial domain to the channel domain for the first time, achieving 100% codebook utilization through a large-scale codebook. Building upon CVQ, they introduce a Channel Autoregressive model (CAR) that employs a “next-channel prediction” mechanism to emulate the human-like coarse-to-fine image generation process. This framework substantially improves image reconstruction fidelity and achieves state-of-the-art performance in text-to-image generation, attaining a DPG score of 86.7 and a GenEval score of 0.79.

Abstract

We present Channel-wise Vector Quantization (CVQ), a novel image tokenization paradigm that replaces patch-wise tokens with channel-wise tokens. Unlike conventional vector quantization, which assigns a discrete token to each patch feature vector, CVQ quantizes each channel of the feature map. This formulation represents an image as discrete levels of visual details, rather than as a grid of spatial patches. Based on CVQ, we introduce a new visual autoregressive framework with "next-channel prediction". Instead of rendering images patch by patch in raster order, our Channel-wise Autoregressive (CAR) model predicts image channels sequentially, producing progressively enriched visual details. Specifically, it first sketches global structure and then refines fine-grained attributes, akin to a human artist's workflow. Empirically, we show that: (1) CVQ achieves 100% codebook utilization with a 16K+ codebook size without any bells and whistles, and substantially improves reconstruction quality over conventional VQ; and (2) CAR attains a DPG score of 86.7 and a GenEval score of 0.79, demonstrating strong effectiveness for text-to-image generation.

Problem

• Vector Quantization
• Image Tokenization
• Channel-wise Representation
• Autoregressive Modeling
• Codebook Utilization

Innovation

• Channel-wise Vector Quantization
• Visual Autoregressive Modeling
• Next-channel Prediction
• Codebook Utilization
• Text-to-Image Generation