Saved in:
Bibliographic Details
Main Authors: Zhu, Xiaoxu, Yu, Xiaojie, Yao, Guangchao, Ren, Yiming, Li, Baoxiang
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.15860
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912992878133248
author Zhu, Xiaoxu
Yu, Xiaojie
Yao, Guangchao
Ren, Yiming
Li, Baoxiang
author_facet Zhu, Xiaoxu
Yu, Xiaojie
Yao, Guangchao
Ren, Yiming
Li, Baoxiang
contents Finite Scalar Quantization (FSQ) offers simplified training but suffers from residual magnitude decay in multi-stage settings, where subsequent stages receive exponentially weaker signals. We propose Robust Residual Finite Scalar Quantization (RFSQ), addressing this fundamental limitation through two novel conditioning strategies: learnable scaling factors and invertible layer normalization. Our experiments across audio and image modalities demonstrate RFSQ's effectiveness and generalizability. In audio reconstruction at 24 bits/frame, RFSQ-LayerNorm achieves 3.646 DNSMOS, a 3.6% improvement over state-of-the-art RVQ (3.518). On ImageNet, RFSQ achieves 0.102 L1 loss and 0.100 perceptual loss, with LayerNorm providing 9.7% L1 improvement and 17.4% perceptual improvement over unconditioned variants. The LayerNorm strategy consistently outperforms alternatives by maintaining normalized input statistics across stages, effectively preventing exponential magnitude decay that limits naive residual approaches. RFSQ combines FSQ's simplicity with multi-stage quantization's representational power, establishing a new standard for neural compression across diverse modalities.
format Preprint
id arxiv_https___arxiv_org_abs_2508_15860
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Robust Residual Finite Scalar Quantization for Neural Compression
Zhu, Xiaoxu
Yu, Xiaojie
Yao, Guangchao
Ren, Yiming
Li, Baoxiang
Image and Video Processing
Computer Vision and Pattern Recognition
Audio and Speech Processing
Finite Scalar Quantization (FSQ) offers simplified training but suffers from residual magnitude decay in multi-stage settings, where subsequent stages receive exponentially weaker signals. We propose Robust Residual Finite Scalar Quantization (RFSQ), addressing this fundamental limitation through two novel conditioning strategies: learnable scaling factors and invertible layer normalization. Our experiments across audio and image modalities demonstrate RFSQ's effectiveness and generalizability. In audio reconstruction at 24 bits/frame, RFSQ-LayerNorm achieves 3.646 DNSMOS, a 3.6% improvement over state-of-the-art RVQ (3.518). On ImageNet, RFSQ achieves 0.102 L1 loss and 0.100 perceptual loss, with LayerNorm providing 9.7% L1 improvement and 17.4% perceptual improvement over unconditioned variants. The LayerNorm strategy consistently outperforms alternatives by maintaining normalized input statistics across stages, effectively preventing exponential magnitude decay that limits naive residual approaches. RFSQ combines FSQ's simplicity with multi-stage quantization's representational power, establishing a new standard for neural compression across diverse modalities.
title Robust Residual Finite Scalar Quantization for Neural Compression
topic Image and Video Processing
Computer Vision and Pattern Recognition
Audio and Speech Processing
url https://arxiv.org/abs/2508.15860