Rollout-Marginal Distillation for Long-Horizon Autoregressive Video Generation
- 类型：arxiv
- 标识：2609.37925
- 链接：https://arxiv.org/abs/2609.37925
- 主分类：multimodal
- 形态：method
- TLDR：Autoregressive (AR) video diffusion enables low-latency, streamable video generation, but prediction errors often accumulate over long rollouts. Training the generator on its own rollouts exposes it to these imperfect histories. However, existing video-level distribution matching distillation (DMD) scores the whole rollout jointly. Because a chunk is evaluated together with its past and future, its correction can favor matching artifacts in the surrounding context merely to preserve temporal consistency. To provide a clearer visual-quality signal, we introduce Rollout-Marginal Distillation (RM
- 待LLM分类：否
- 标题中文：Rollout-Marginal Distillation for Long-Horizon Autoregressive Video Generation：面向长程自回归视频生成的 Rollout-Marginal 蒸馏
- TLDR中文：[TLDR -> TLDR中文] 自回归 (AR) 视频扩散支持低延迟、可流式生成的视频，但预测误差在长时序展开过程中会不断累积。让生成器在其自身展开结果上进行训练，可以使其接触这些不完美的历史。然而，现有的视频级分布匹配蒸馏 (DMD) 对整个展开序列进行联合打分。由于一个片段与其前后片段一起被评估，其校正信号可能仅仅为了维持时间一致性而倾向于匹配周围上下文中的伪影。为了提供更清晰的视觉质量信号，我们提出 Rollout-Marginal Distillation (RM
- 来源文件：
- /inbox/tom/_candidates/2026-10-05-agent-rag-longcontext-candidates.json