Quantize-then-Rectify: Efficient VQ-VAE Training

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Borui, Rao, Qihang, Zheng, Wenzhao, Zhou, Jie, Lu, Jiwen
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911054981758976
author Zhang, Borui
Rao, Qihang
Zheng, Wenzhao
Zhou, Jie
Lu, Jiwen
author_facet Zhang, Borui
Rao, Qihang
Zheng, Wenzhao
Zhou, Jie
Lu, Jiwen
contents Visual tokenizers are pivotal in multimodal large models, acting as bridges between continuous inputs and discrete tokens. Nevertheless, training high-compression-rate VQ-VAEs remains computationally demanding, often necessitating thousands of GPU hours. This work demonstrates that a pre-trained VAE can be efficiently transformed into a VQ-VAE by controlling quantization noise within the VAE's tolerance threshold. We present \textbf{Quantize-then-Rectify (ReVQ)}, a framework leveraging pre-trained VAEs to enable rapid VQ-VAE training with minimal computational overhead. By integrating \textbf{channel multi-group quantization} to enlarge codebook capacity and a \textbf{post rectifier} to mitigate quantization errors, ReVQ compresses ImageNet images into at most 512 tokens while sustaining competitive reconstruction quality (rFID = 1.06). Significantly, ReVQ reduces training costs by over two orders of magnitude relative to state-of-the-art approaches: ReVQ finishes full training on a single NVIDIA 4090 in approximately 22 hours, whereas comparable methods require 4.5 days on 32 A100 GPUs. Experimental results show that ReVQ achieves superior efficiency-reconstruction trade-offs.
format Preprint
id arxiv_https___arxiv_org_abs_2507_10547
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Quantize-then-Rectify: Efficient VQ-VAE Training
Zhang, Borui
Rao, Qihang
Zheng, Wenzhao
Zhou, Jie
Lu, Jiwen
Computer Vision and Pattern Recognition
Machine Learning
Visual tokenizers are pivotal in multimodal large models, acting as bridges between continuous inputs and discrete tokens. Nevertheless, training high-compression-rate VQ-VAEs remains computationally demanding, often necessitating thousands of GPU hours. This work demonstrates that a pre-trained VAE can be efficiently transformed into a VQ-VAE by controlling quantization noise within the VAE's tolerance threshold. We present \textbf{Quantize-then-Rectify (ReVQ)}, a framework leveraging pre-trained VAEs to enable rapid VQ-VAE training with minimal computational overhead. By integrating \textbf{channel multi-group quantization} to enlarge codebook capacity and a \textbf{post rectifier} to mitigate quantization errors, ReVQ compresses ImageNet images into at most 512 tokens while sustaining competitive reconstruction quality (rFID = 1.06). Significantly, ReVQ reduces training costs by over two orders of magnitude relative to state-of-the-art approaches: ReVQ finishes full training on a single NVIDIA 4090 in approximately 22 hours, whereas comparable methods require 4.5 days on 32 A100 GPUs. Experimental results show that ReVQ achieves superior efficiency-reconstruction trade-offs.
title Quantize-then-Rectify: Efficient VQ-VAE Training
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2507.10547