让视频世界模型更快推理，减少模糊和拖影。
WorldCache: Content-Aware Caching for Accelerated Video World Models
- 根据运动动态调整缓存时机与方式
- 速度提升2.3倍，质量损失仅0.6%
- 无需重新训练，适合实时视频生成
扩散变换器（DiTs）驱动高保真视频世界模型，但因逐步去噪和高代价时空注意力而计算开销大。无训练特征缓存通过复用中间激活加速推理，但现有方法多依赖零阶保持假设——在全局漂移较小时将缓存特征视为静态快照，常导致动态场景中出现鬼影、模糊和运动不一致。本文提出感知约束的动力学缓存框架WorldCache，引入运动自适应阈值、显著性加权漂移估计、最优混合与扭曲逼近及扩散步骤阶段感知的阈值调度。该协同方法实现无需重训练的自适应、运动一致特征复用。在PAI-Bench上的Cosmos-Predict2.5-2B数据集上，WorldCache实现2.3倍推理加速，同时保持99.4%的基线质量，显著优于以往无训练缓存方法。
原文摘要 · Abstract (English)
Diffusion Transformers (DiTs) power high-fidelity video world models but remain computationally expensive due to sequential denoising and costly spatio-temporal attention. Training-free feature caching accelerates inference by reusing intermediate activations across denoising steps; however, existing methods largely rely on a Zero-Order Hold assumption i.e., reusing cached features as static snapshots when global drift is small. This often leads to ghosting artifacts, blur, and motion inconsistencies in dynamic scenes. We propose \textbf{WorldCache}, a Perception-Constrained Dynamical Caching framework that improves both when and how to reuse features. WorldCache introduces motion-adaptive thresholds, saliency-weighted drift estimation, optimal approximation via blending and warping, and phase-aware threshold scheduling across diffusion steps. Our cohesive approach enables adaptive, motion-consistent feature reuse without retraining. On Cosmos-Predict2.5-2B evaluated on PAI-Bench, WorldCache achieves \textbf{2.3$\times$} inference speedup while preserving \textbf{99.4\%} of baseline quality, substantially outperforming prior training-free caching approaches. Our code can be accessed on \href{https://umair1221.github.io/World-Cache/}{World-Cache}.
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。