提出球面量化方法，让视觉表示更稳定可扩展。
dRAE: Representation Autoencoder with Hyper-Spherical Codes
- 用角度路由分离语义与幅值，避免量化解码器崩溃。
- 代码本规模达13.1万时仍保持100%利用率和高保真重建。
- 适合需要大规模离散表示的视觉-语言任务研究者。
本文旨在将高维视觉表示离散化，以弥合与语言模型的差距——这是一项非平凡挑战，因现有量化方法存在代码本崩溃问题，难以在保持语义连贯性的同时扩展。我们发现根源在于度量不匹配：标准欧氏代码本目标与表示空间的各向异性几何本质不符，导致代码本嵌入幅值变化大、角度分布不均，阻碍可扩展性。为此，我们提出超球面量化（HSQ），通过角度路由解耦语义内容与特征幅值，防止代码分配被尺度主导而偏离语义。由此构建的离散表示自编码器（dRAE）实现高保真重建，同时保持语义完整性并支持可扩展的代码本预算。大量实验表明，当词汇表规模扩展至131,072时，性能持续提升，代码本利用率达100%，训练流程简化，并在理解与生成任务中表现强劲。
原文摘要 · Abstract (English)
In this work, we aim to discretize the high-dimensional visual representations to bridge the gap with language models - a non-trivial challenge, as existing quantization methods suffer from codebook collapse, failing to scale while preserving semantic coherence. We identify the root cause as metric mismatch: standard Euclidean codebook objectives are fundamentally misaligned with the anisotropic geometry of representation space, leading to codebook embeddings with high-variance magnitude scales and uneven angular distributions that hinder scalability. To address this, we propose Hyper-Spherical Quantization (HSQ), which decouples semantic content from feature magnitude via angular routing, preventing code assignment from being dominated by scale rather than meaning. The resulting discrete Representation Autoencoder (dRAE) achieves high-fidelity reconstruction while preserving semantic integrity and supporting scalable codebook budget. Extensive experiments demonstrate consistent performance gains as the vocabulary size scales to 131{,}072, along with 100\% codebook utilization, simplified training pipeline, and strong performance across understanding and generation tasks.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。