Saved in:
Bibliographic Details
Main Authors: Zhao, Yue, Jiang, Hanwen, Xu, Zhenlin, Yang, Chutong, Adeli, Ehsan, Krähenbühl, Philipp
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.14697
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915679487131648
author Zhao, Yue
Jiang, Hanwen
Xu, Zhenlin
Yang, Chutong
Adeli, Ehsan
Krähenbühl, Philipp
author_facet Zhao, Yue
Jiang, Hanwen
Xu, Zhenlin
Yang, Chutong
Adeli, Ehsan
Krähenbühl, Philipp
contents Non-parametric quantization has received much attention due to its efficiency on parameters and scalability to a large codebook. In this paper, we present a unified formulation of different non-parametric quantization methods through the lens of lattice coding. The geometry of lattice codes explains the necessity of auxiliary loss terms when training auto-encoders with certain existing lookup-free quantization variants such as BSQ. As a step forward, we explore a few possible candidates, including random lattices, generalized Fibonacci lattices, and densest sphere packing lattices. Among all, we find the Leech lattice-based quantization method, which is dubbed as Spherical Leech Quantization ($Λ_{24}$-SQ), leads to both a simplified training recipe and an improved reconstruction-compression tradeoff thanks to its high symmetry and even distribution on the hypersphere. In image tokenization and compression tasks, this quantization approach achieves better reconstruction quality across all metrics than BSQ, the best prior art, while consuming slightly fewer bits. The improvement also extends to state-of-the-art auto-regressive image generation frameworks.
format Preprint
id arxiv_https___arxiv_org_abs_2512_14697
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Spherical Leech Quantization for Visual Tokenization and Generation
Zhao, Yue
Jiang, Hanwen
Xu, Zhenlin
Yang, Chutong
Adeli, Ehsan
Krähenbühl, Philipp
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Signal Processing
Non-parametric quantization has received much attention due to its efficiency on parameters and scalability to a large codebook. In this paper, we present a unified formulation of different non-parametric quantization methods through the lens of lattice coding. The geometry of lattice codes explains the necessity of auxiliary loss terms when training auto-encoders with certain existing lookup-free quantization variants such as BSQ. As a step forward, we explore a few possible candidates, including random lattices, generalized Fibonacci lattices, and densest sphere packing lattices. Among all, we find the Leech lattice-based quantization method, which is dubbed as Spherical Leech Quantization ($Λ_{24}$-SQ), leads to both a simplified training recipe and an improved reconstruction-compression tradeoff thanks to its high symmetry and even distribution on the hypersphere. In image tokenization and compression tasks, this quantization approach achieves better reconstruction quality across all metrics than BSQ, the best prior art, while consuming slightly fewer bits. The improvement also extends to state-of-the-art auto-regressive image generation frameworks.
title Spherical Leech Quantization for Visual Tokenization and Generation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Signal Processing
url https://arxiv.org/abs/2512.14697