Quantize-then-Rectify: Efficient VQ-VAE Training

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 $\mathbf{\text{Quantize-then-Rectify}}\mathbf{\text{(ReVQ)}}$, a framework leveraging pre-trained VAEs to enable rapid VQ-VAE training with minimal computational overhead. By integrating $\mathbf{\text{channel}}\mathbf{\text{multi-group}}\mathbf{\text{quantization}}$ to enlarge codebook capacity and a $\mathbf{\text{post}}\mathbf{\text{rectifier}}$ 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.