立方离散扩散：基于高维表征词元的离散视觉生成
Cubic Discrete Diffusion: Discrete Visual Generation on High-Dimensional Representation Tokens
把预训练高维语义表征（768-1024 维）直接离散化后做掩码扩散生成，使同一套离散词元既能用于理解又能用于生成，为视觉生成与语言模型共享词元预测范式提供了可复用的路径；其「任意维度任意位置可掩码、生成步数固定为 T」的机制可直接迁移到高维特征上的生成式建模。
原文依据 CubiD performs fine-grained masking throughout the high-dimensional discrete representation--any dimension at any position can be masked and predicted from partial observations.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论基于离散词元的视觉生成已获得广泛关注，因为它能够与语言模型共享统一的词元预测范式，有望实现无缝的多模态架构。然而，当前的离散生成方法仍局限于低维潜变量词元（通常为 8-32 维），牺牲了对理解至关重要的语义丰富性。尽管高维预训练表征（768-1024 维）可以弥合这一差距，但其离散生成面临根本性挑战。本文提出立方离散扩散（CubiD），这是首个面向高维表征的离散生成模型。CubiD 在整个高维离散表征上进行细粒度掩码——任意位置上的任意维度都可以被掩码，并由部分观测进行预测。这使得模型能够学习空间位置内部以及跨空间位置的丰富关联，且生成步数固定为 T，与特征维度无关，其中 T << hwd。在 ImageNet-256 上，CubiD 取得了最优的离散生成性能，并在 900M 到 3.7B 参数范围内展现出良好的扩展行为。关键在于，我们验证了这些离散化词元保留了原始表征能力，表明同一套离散词元可以同时有效地服务于理解与生成任务。我们希望这项工作能够启发未来面向统一多模态架构的研究。代码已发布于：https://github.com/YuqingWang1029/CubiD。
英文原文摘要 1510 字符
Visual generation with discrete tokens has gained significant attention as it enables a unified token prediction paradigm shared with language models, promising seamless multimodal architectures. However, current discrete generation methods remain limited to low-dimensional latent tokens (typically 8-32 dims), sacrificing the semantic richness essential for understanding. While high-dimensional pretrained representations (768-1024 dims) could bridge this gap, their discrete generation poses fundamental challenges. In this paper, we present Cubic Discrete Diffusion (CubiD), the first discrete generation model for high-dimensional representations. CubiD performs fine-grained masking throughout the high-dimensional discrete representation--any dimension at any position can be masked and predicted from partial observations. This enables the model to learn rich correlations both within and across spatial positions, with the number of generation steps fixed at T regardless of feature dimensionality, where T << hwd. On ImageNet-256, CubiD achieves state-of-the-art discrete generation with strong scaling behavior from 900M to 3.7B parameters. Crucially, we validate that these discretized tokens preserve original representation capabilities, demonstrating that the same discrete tokens can effectively serve both understanding and generation tasks. We hope this work will inspire future research toward unified multimodal architectures. Code is available at: https://github.com/YuqingWang1029/CubiD.