多步驟視覺推理的層級式去噪

摘要

視頻模型正逐漸演變為視覺基礎模型，但仍缺乏人類般的多步驟推理能力。串流自迴歸擴散模型雖具效率，但在推理上有所局限；而雙向擴散雖能實現全局修正，卻因密集的逐幀去噪而導致高昂的推論成本。這兩種典範在處理複雜推理任務時，皆難以兼顧邏輯一致性與低延遲串流。我們提出 HDR（分層去噪視覺推理），這是一個統一框架，將分層潛變量整合至因果式影片生成中，以實現多步驟推理。HDR 將影片潛變量組織成樹狀層級結構，能夠在串流輸出前先進行由粗到細的推理。粗層去噪保留不確定假設以利全局規劃，而細層則逐步將這些假設精煉為具體的視覺狀態。此外，稀疏分層注意力模式（SHAP）進一步降低了時間注意力成本。我們引入一個包含分布外案例的層級分層多步驟影片推理基準，涵蓋六項任務：迷宮導航、河內塔、一筆畫、滑塊拼圖、倉庫番及倒水問題。與串流自迴歸擴散基線相比，HDR 將成功率從 34.22 提升至 60.29（相對增益 76.2%），並將平均進度從 76.00 提升至 89.56，展現更一致的推理軌跡。HDR 維持每個潛變量 0.70 秒的低延遲串流，推論速度比雙向擴散快 54.2 倍。此外，HDR 在僅使用 2% 訓練資料的情況下，仍保留 82.9% 的完整資料效能，而雙向擴散僅為 52.0%。實際機器人實驗更進一步證明 HDR 在物理互動與世界模型方面的潛力。專案展示：https://hierarchical-diffusion-reasoning.github.io/。

English

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 layers progressively refine them into concrete visual states. A sparse hierarchical attention pattern (SHAP) further reduces temporal attention costs. We introduce a level-stratified multi-step video reasoning benchmark with out-of-distribution cases, covering six tasks: maze navigation, Tower of Hanoi, one-line drawing, sliding puzzle, Sokoban, and water pouring. Compared with streaming autoregressive diffusion baselines, HDR improves success from 34.22 to 60.29 (76.2% relative gain) and increases average progress from 76.00 to 89.56, demonstrating more consistent reasoning trajectories. HDR maintains low-latency streaming at 0.70 seconds per latent, achieving 54.2 times faster inference than bidirectional diffusion. It also retains 82.9% of full-data performance with only 2% training data, compared with 52.0% for bidirectional diffusion. Real-world robot experiments further demonstrate HDR's potential for physical interaction and world modeling. Project demo: https://hierarchical-diffusion-reasoning.github.io/.