English(EN) GraftOOD: Grafting Out-of-Distribution Exemplars into Advantage-Weighted Rectification for Video Reward Optimization

GraftOOD 方法通过嵌入分布外样本来增强视频奖励优化

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-12 04:00

研究人员推出了一种新颖的视频奖励优化方法 GraftOOD，通过引入分布外视频样本来解决“运动崩溃”问题。该技术用这些外部样本替换一些在线采样，旨在在不牺牲运动质量的情况下改善视频动态。GraftOOD 在 Wan2.1-1.3B 模型上进行了测试，显示动态度显著提高，同时保持或改善了其他指标，优于现有的样本增强方法。 AI

影响 GraftOOD 提供了一种新的方法来提高视频生成质量，通过平衡动态和运动连贯性，可能影响未来的视频合成模型。

排序理由该集群包含一篇研究论文，详细介绍了一种新的视频奖励优化方法。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CV 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 10 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇研究论文，详细介绍了一种新的视频奖励优化方法。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, other; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Same-day; Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CV TIER_1 English(EN) · Zheming Fu, Ruizhe He, Wenbin Ning, Xiaoxiao Ma, Xuan Yang, Lianrui Mu, Haoyang Huang, Chang Liu, Lei Wang, Siming Fu · 2026-10-12 04:00

GraftOOD：将分布外样本嫁接到优势加权校正以优化视频奖励

arXiv:2610.13470v1 Announce Type: new Abstract: Video reward optimization can increase proxy rewards while reducing video dynamics, a mismatch we identify as \emph{motion collapse}, including under a video-level motion-quality reward. We propose \textbf{GraftOOD}, which replaces …

本文如何被排名

报道来源 [1]

GraftOOD：将分布外样本嫁接到优势加权校正以优化视频奖励

相关实体

相关话题