Spherical Leech Quantization for Visual Tokenization and Generation
This work addresses efficiency and performance in visual tokenization and generation, offering an incremental improvement over existing non-parametric quantization methods.
The paper tackled the problem of non-parametric quantization in visual tasks by proposing Spherical Leech Quantization (Λ24-SQ), which improved reconstruction quality over prior methods like BSQ while using slightly fewer bits in image tokenization and compression.
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.