[论文] Hierarchical Denoising For Multi-Step Visual Reasoning

论文概要

研究领域: CV
作者: Zezhong Qian, Xiaowei Chi, Chak-Wing Mak, Tianze Zhou, Ruibin Yuan, Yuhan Rui, Hengzhe Sun, Zhuoqun Wu, Yuming Li, Siyuan Qian, Sirui Han, Shanghang Zhang
发布时间: 2026-07-16
arXiv: 2607.15278

中文摘要

视频模型正在演变为视觉基础模型，但仍缺乏类人化的多步推理能力。流式自回归扩散模型效率虽高，但推理能力受限；双向扩散虽能实现全局修订，却因逐帧密集去噪而推理成本高昂。这两种范式都难以在复杂推理任务中同时实现逻辑一致性和低延迟流式输出。本文提出HDR（面向视觉推理的分层去噪），一种将分层潜变量整合到因果视频生成中的统一框架，用于多步推理。HDR将视频潜变量组织为树状层次结构，在流式输出前实现从粗到精的推理。粗去噪层保留不确定的假设以进行全局规划，而精细层逐步将其细化为具体的视觉状态。一种稀疏分层注意力机制（SHAP）进一步降低了时间注意力成本。本文还引入了一个包含分布外案例的分层多步视频推理基准，涵盖六项任务：迷宫导航、汉诺塔、单线绘制、滑块拼图、推箱子和倒水。与流式自回归扩散基线相比，HDR将成功率从34.22提升至60.29（相对提升76.2%），平均进度从76.00提升至89.56，展现出更一致的推理轨迹。HDR保持低延迟流式输出，每潜变量仅需0.70秒，推理速度比双向扩散快54.2倍。在仅使用2%训练数据的情况下，HDR仍保持了全量数据82.9%的性能，而双向扩散仅能保持52.0%。真实世界机器人实验进一步展示了HDR在物理交互和世界建模方面的潜力。

原文摘要

Video models are evolving into vision foundation models, yet they still lack human-like multi-step reasoning. Streaming autoregressive diffusion models are efficient but limited in reasoning, while bidirectional diffusion enables global revision with high inference costs due to dense frame-level denoising. Both paradigms struggle to achieve logical consistency and low-latency streaming for complex reasoning tasks. We propose HDR (Hierarchical Denoising for Visual Reasoning), a unified framework that integrates hierarchical latents into causal video generation for multi-step reasoning. HDR organizes video latents into a tree-structured hierarchy, enabling coarse-to-fine reasoning before streaming output. Coarse denoising layers preserve uncertain hypotheses for global planning, while finer ...

自动采集于 2026-07-18

#论文 #arXiv #CV #小凯