因果强迫：自回归扩散蒸馏的正确方法，用于高质量实时交互式视频生成
Causal Forcing: Autoregressive Diffusion Distillation Done Right for High-Quality Real-Time Interactive Video Generation
AI总结：
针对双向扩散模型蒸馏为自回归模型时的架构差距问题，提出因果强迫方法，通过自回归教师进行ODE初始化并应用DMD过程，显著提升视频生成质量。
摘要：
为了实现实时交互式视频生成，当前方法将预训练的双向视频扩散模型蒸馏为少步自回归（AR）模型，当全注意力被因果注意力替代时面临架构差距。然而，现有方法并未从理论上弥合这一差距。它们通过ODE蒸馏初始化AR学生模型，这需要帧级单射性，即在AR教师的PF-ODE下，每个噪声帧必须映射到唯一的干净帧。从双向教师蒸馏AR学生违反了这一条件，阻止了教师流映射的恢复，反而诱导出条件期望解，导致性能下降。为解决此问题，我们提出因果强迫（Causal Forcing），它使用自回归教师进行ODE初始化以弥合架构差距，然后应用与Self Forcing相同的DMD过程。实验结果表明，我们的方法在所有指标上优于所有基线，在动态程度、VisionReward和指令跟随上分别超过SOTA Self Forcing 19.3%、8.7%和16.7%。项目页面：https://thu-ml.github.io/CausalForcing.github.io/；代码：https://github.com/thu-ml/Causal-Forcing。
英文摘要：
To achieve real-time interactive video generation, current methods distill pretrained bidirectional video diffusion models into few-step autoregressive (AR) models, facing an architectural gap when full attention is replaced by causal attention. However, existing approaches do not bridge this gap theoretically. They initialize the AR student via ODE distillation, which requires frame-level injectivity, where each noisy frame must map to a unique clean frame under the PF-ODE of an AR teacher. Distilling an AR student from a bidirectional teacher violates this condition, preventing recovery of the teacher's flow map and instead inducing a conditional-expectation solution, which degrades performance. To address this issue, we propose Causal Forcing, which uses an autoregressive teacher for ODE initialization to bridge the architectural gap, and then applies the same DMD procedure as in Self Forcing. Empirical results show that our method outperforms all baselines across all metrics, surpassing the SOTA Self Forcing by 19.3\% in Dynamic Degree, 8.7\% in VisionReward, and 16.7\% in Instruction Following. Project page: \href{https://thu-ml.github.io/CausalForcing.github.io/}{https://thu-ml.github.io/CausalForcing.github.io/}; the code: \href{https://github.com/thu-ml/Causal-Forcing}{https://github.com/thu-ml/Causal-Forcing}.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
