Spherical Leech Quantization for Visual Tokenization and Generation

2025/12/16 by Yue Zhao, Hanwen Jiang, Zhao, Yue +8
Computer Science · #Advanced Data Compression Techniques #Advanced Image and Video Retrieval Techniques #Artificial Intelligence (cs.AI) #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Generative Adversarial Networks and Image Synthesis #Machine Learning (cs.LG) #Signal Processing (eess.SP) #electronic engineering #information engineering

paper · pdf · doi:10.48550/arxiv.2512.14697

openalex publication_date 2025/12/16 · openalex created_date 2025/12/18 · openalex updated_date 2026/07/28

Abstract

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 (Λ₂₄-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.

Citations

• AToken: A Unified Tokenizer for Vision
• GigaTok: Scaling Visual Tokenizers to 3 Billion Parameters for Autoregressive Image Generation
• Dion: Distributed Orthonormalized Updates
• UniTok: A Unified Tokenizer for Visual Generation and Understanding
• Muon is Scalable for LLM Training
• QLIP: Text-Aligned Visual Tokenization Unifies Auto-Regressive Multimodal Understanding and Generation
• DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning
• Learnings from Scaling Visual Tokenizers for Reconstruction and Generation
• Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models
• 2 OLMo 2 Furious
• When Worse is Better: Navigating the compression-generation tradeoff in visual tokenization
• Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis
• Scalable Image Tokenization with Index Backpropagation Quantization
• RandAR: Decoder-only Autoregressive Visual Generation in Random Orders
• Cut Your Losses in Large-Vocabulary Language Models
• Randomized Autoregressive Visual Generation
• GPT-4o System Card
• Restructuring Vector Quantization with the Rotation Trick
• nGPT: Normalized Transformer with Representation Learning on the Hypersphere
• MaskBit: Embedding-free Image Generation via Bit Tokens
• Scaling Laws with Vocabulary: Larger Models Deserve Larger Vocabularies
• OmniTokenizer: A Joint Image-Video Tokenizer for Visual Generation
• Image and Video Tokenization with Binary Spherical Quantization
• An Image is Worth 32 Tokens for Reconstruction and Generation
• Autoregressive Model Beats Diffusion: Llama for Scalable Image Generation
• Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction
• Photorealistic Video Generation with Diffusion Models
• Sequential Modeling Enables Scalable Learning for Large Vision Models
• Language Model Beats Diffusion -- Tokenizer is Key to Visual Generation
• Finite Scalar Quantization: VQ-VAE Made Simple
• Small-scale proxies for large-scale Transformer training instabilities
• Language Modeling Is Compression
• Straightening Out the Straight-Through Estimator: Overcoming Optimization Challenges in Vector Quantized Networks
• DINOv2: Learning Robust Visual Features without Supervision
• Symbolic Discovery of Optimization Algorithms
• Classifier-Free Diffusion Guidance
• PaLM: Scaling Language Modeling with Pathways
• Autoregressive Image Generation using Residual Quantization
• MaskGIT: Masked Generative Image Transformer
• Self-supervised Learning with Random-projection Quantizer for Speech Recognition
• High-Resolution Image Synthesis with Latent Diffusion Models
• Vector-quantized Image Modeling with Improved VQGAN
• BEiT: BERT Pre-Training of Image Transformers
• Diffusion Models Beat GANs on Image Synthesis
• Zero-Shot Text-to-Image Generation
• Taming Transformers for High-Resolution Image Synthesis
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• Language Models are Few-Shot Learners
• Coincidence, Categorization, and Consolidation: Learning to Recognize Sounds with Minimal Supervision
• vq-wav2vec: Self-Supervised Learning of Discrete Speech Representations
• The Curious Case of Neural Text Degeneration
• Improved Precision and Recall Metric for Assessing Generative Models
• The Unreasonable Effectiveness of Deep Features as a Perceptual Metric
• Neural Discrete Representation Learning
• Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representations
• Conditional Image Generation with PixelCNN Decoders
• Improved Techniques for Training GANs
• Daala: A Perceptually-Driven Next Generation Video Codec
• ImageNet Large Scale Visual Recognition Challenge
• Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation

Related

• Scaling Image Tokenizers with Grouped Spherical Quantization
• Image and Video Tokenization with Binary Spherical Quantization
• Improving 3D Gaussian Splatting Compression by Scene-Adaptive Lattice Vector Quantization
• LL-VQ-VAE: Learnable Lattice Vector-Quantization For Efficient Representations
• LVQAC: Lattice Vector Quantization Coupled with Spatially Adaptive Companding for Efficient Learned Image Compression