Papers › MGVQ: Could VQ-VAE Beat VAE? A Generalizable Tokenizer with Multi-group Quantization

MGVQ: Could VQ-VAE Beat VAE? A Generalizable Tokenizer with Multi-group Quantization

14 Jul 2025archive 2025-07-28

Mingkai Jia, Wei Yin, Xiaotao Hu, Jiaxin Guo, Xiaoyang Guo, Qian Zhang, Xiao-Xiao Long, Ping Tan

Vector Quantized Variational Autoencoders (VQ-VAEs) are fundamental models that compress continuous visual data into discrete tokens. Existing methods have tried to improve the quantization strategy for better reconstruction quality, however, there still exists a large gap between VQ-VAEs and VAEs. To narrow this gap, we propose MGVQ, a novel method to augment the representation capability of discrete codebooks, facilitating easier optimization for codebooks and minimizing information loss, thereby enhancing reconstruction quality. Specifically, we propose to retain the latent dimension to preserve encoded features and incorporate a set of sub-codebooks for quantization. Furthermore, we construct comprehensive zero-shot benchmarks featuring resolutions of 512p and 2k to evaluate the reconstruction performance of existing methods rigorously. MGVQ achieves the state-of-the-art performance on both ImageNet and 8 zero-shot benchmarks across all VQ-VAEs. Notably, compared with SD-VAE, we outperform them on ImageNet significantly, with rFID 0.49 v.s. 0.91, and achieve superior PSNR on all zero-shot benchmarks. These results highlight the superiority of MGVQ in reconstruction and pave the way for preserving fidelity in HD image processing tasks. Code will be publicly available at https://github.com/MKJia/MGVQ

PaperPDFCode

Code

MKJia/MGVQ mentioned in paper report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

2kImage GenerationImage ReconstructionQuantization

Datasets

Introduced by this paper, per the archive.

Ultra-High Resolution Image Reconstruction Benchmark

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Image Generation ImageNet 256x256 MGVQ FID 3.02 #66 of 94 Archive leaderboard report
Image Generation ImageNet 256x256 MGVQ Inception score 294.1 #66 of 94 Archive leaderboard report
Image Reconstruction ImageNet MGVQ (16x16x8) FID 0.49 #1 of 15 Archive leaderboard report
Image Reconstruction ImageNet MGVQ (16x16x8) LPIPS 0.086 #1 of 15 Archive leaderboard report
Image Reconstruction ImageNet MGVQ (16x16x8) PSNR 24.70 #1 of 15 Archive leaderboard report
Image Reconstruction ImageNet MGVQ (16x16x8) SSIM 0.787 #1 of 15 Archive leaderboard report
Image Reconstruction ImageNet MGVQ (16x16x4) FID 0.64 #2 of 15 Archive leaderboard report
Image Reconstruction ImageNet MGVQ (16x16x4) LPIPS 0.110 #2 of 15 Archive leaderboard report
Image Reconstruction ImageNet MGVQ (16x16x4) PSNR 23.71 #2 of 15 Archive leaderboard report
Image Reconstruction ImageNet MGVQ (16x16x4) SSIM 0.755 #2 of 15 Archive leaderboard report
Image Reconstruction Ultra-High Resolution Image Reconstruction Benchmark MGVQ (16x16x4) LPIPS 0.092 #2 of 6 Archive leaderboard report
Image Reconstruction Ultra-High Resolution Image Reconstruction Benchmark MGVQ (16x16x4) PSNR 28.27 #2 of 6 Archive leaderboard report
Image Reconstruction Ultra-High Resolution Image Reconstruction Benchmark MGVQ (16x16x4) SSIM 0.844 #2 of 6 Archive leaderboard report
Image Reconstruction Ultra-High Resolution Image Reconstruction Benchmark MGVQ (16x16x4) rFID 1.59 #2 of 6 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections