Distributional Matching for Vector Quantization: A Unified Theoretical and Empirical Framework
arXiv, Under Review
Xianghong Fang, Litao Guo, Hengchao Chen, Yuxuan Zhang, Xiaofan Xia, Dingjie Song, Yexin Liu, Hao Wang, Harry Yang, Qiang Sun, Yuan Yuan
TL;DR Matching feature and codebook distributions provides a unified approach to reducing vector-quantization instability and codebook collapse, with Wasserstein and maximum mean discrepancy objectives supported by theoretical analysis and visual-tokenization experiments.