VibeToken：扩展一维图像分词器与自回归模型以实现动态分辨率生成
VibeToken: Scaling 1D Image Tokenizers and Autoregressive Models for Dynamic Resolution Generations
把图像编码成 32–256 个可动态控制长度的一维 token，使自回归生成器在任意分辨率下保持恒定浮点运算量；做视觉生成时可借鉴其分辨率无关的分词器设计，替代固定分辨率 VQ 分词器来降低高分辨率推理成本。
原文依据 VibeToken-Gen maintains a constant 179G FLOPs (63.4x efficient) independent of resolution
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论我们提出一种高效、分辨率无关的自回归（AR）图像合成方法，可泛化到任意分辨率和宽高比，在大规模下缩小了与扩散模型的差距。其核心是 VibeToken，一种新颖的分辨率无关一维 Transformer 图像分词器，可将图像编码为动态、用户可控的 32–256 个 token 序列，实现了最优的效率与性能权衡。基于 VibeToken，我们提出 VibeToken-Gen，一个类别条件自回归生成器，开箱即支持任意分辨率，同时所需计算资源显著更少。值得注意的是，VibeToken-Gen 仅用 64 个 token 即可合成 1024x1024 图像，并达到 3.94 gFID；相比之下，基于扩散的最优替代方案需要 1,024 个 token 才能达到 5.87 gFID。与 LlamaGen 等固定分辨率自回归模型——其推理浮点运算量随分辨率呈二次增长（1024x1024 时为 11T 浮点运算量）——不同，VibeToken-Gen 保持恒定的 179G 浮点运算量（效率提升 63.4 倍），且与分辨率无关。我们希望 VibeToken 能推动自回归视觉生成模型在生产场景中的广泛采用。项目主页：https://github.com/SonyResearch/VibeToken
英文原文摘要 1235 字符
We introduce an efficient, resolution-agnostic autoregressive (AR) image synthesis approach that generalizes to arbitrary resolutions and aspect ratios, narrowing the gap to diffusion models at scale. At its core is VibeToken, a novel resolution-agnostic 1D Transformer-based image tokenizer that encodes images into a dynamic, user-controllable sequence of 32-256 tokens, achieving a state-of-the-art efficiency and performance trade-off. Building on VibeToken, we present VibeToken-Gen, a class-conditioned AR generator with out-of-the-box support for arbitrary resolutions while requiring significantly fewer compute resources. Notably, VibeToken-Gen synthesizes 1024x1024 images using only 64 tokens and achieves 3.94 gFID; by comparison, a diffusion-based state-of-the-art alternative requires 1,024 tokens and attains 5.87 gFID. In contrast to fixed-resolution AR models such as LlamaGen - whose inference FLOPs grow quadratically with resolution ( 11T FLOPs at 1024x1024) - VibeToken-Gen maintains a constant 179G FLOPs (63.4x efficient) independent of resolution. We hope VibeToken can help unlock the wide adoption of AR visual generative models in production use cases. Project page: https://github.com/SonyResearch/VibeToken