arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2601.04033cs.CV
收藏

基于框架的思考:通过帧奖励模型生成视频失真评估

Thinking with Frames: Generative Video Distortion Evaluation via Frame Reward Model

University of Science and Technology of China(中国科学技术大学);Kling Team, Kuaishou Technology(快手科技 Kling 团队);Institute of Software Chinese Academy of Sciences(中国科学院软件研究所)

Yuan Wang, Borui Liao, Huijuan Huang, Jinda Lu, Ouxiang Li, Kuien Liu, Meng Wang, Xiang Wang

AI总结:

本文提出REACT框架,通过帧级奖励模型评估生成视频的结构性失真,结合人类偏好数据集和高效合成流程,提升视频生成质量评估的准确性和可解释性。

摘要:

近期视频奖励模型和后训练策略的进步提升了文本到视频(T2V)生成能力。尽管这些模型通常评估视觉质量、运动质量和文本对齐,但往往忽视关键结构性失真,如异常物体外观和交互,这会降低生成视频的整体质量。为解决这一差距,我们引入REACT,一种专门用于生成视频结构性失真评估的帧级奖励模型。REACT通过对视频帧进行推理,分配点评分和归因标签,专注于识别失真。为此,我们构建了一个大规模的人类偏好数据集,基于我们提出的结构性失真分类法进行标注,并使用高效的链式推理(CoT)合成流程生成额外数据。REACT采用两阶段框架:(1)使用掩码损失进行监督微调以注入领域知识,随后(2)使用组相对策略优化(GRPO)和成对奖励进行强化学习,以增强推理能力并使输出评分与人类偏好一致。在推理过程中,引入动态采样机制,以聚焦于最可能表现出失真的帧。我们还提出了REACT-Bench,一个用于生成视频失真评估的基准。实验结果表明,REACT在评估结构性失真方面补充了现有奖励模型,实现了准确的定量评估和可解释的归因分析。

英文摘要:

Recent advances in video reward models and post-training strategies have improved text-to-video (T2V) generation. While these models typically assess visual quality, motion quality, and text alignment, they often overlook key structural distortions, such as abnormal object appearances and interactions, which can degrade the overall quality of the generative video. To address this gap, we introduce REACT, a frame-level reward model designed specifically for structural distortions evaluation in generative videos. REACT assigns point-wise scores and attribution labels by reasoning over video frames, focusing on recognizing distortions. To support this, we construct a large-scale human preference dataset, annotated based on our proposed taxonomy of structural distortions, and generate additional data using a efficient Chain-of-Thought (CoT) synthesis pipeline. REACT is trained with a two-stage framework: (1) supervised fine-tuning with masked loss for domain knowledge injection, followed by (2) reinforcement learning with Group Relative Policy Optimization (GRPO) and pairwise rewards to enhance reasoning capability and align output scores with human preferences. During inference, a dynamic sampling mechanism is introduced to focus on frames most likely to exhibit distortion. We also present REACT-Bench, a benchmark for generative video distortion evaluation. Experimental results demonstrate that REACT complements existing reward models in assessing structutal distortion, achieving both accurate quantitative evaluations and interpretable attribution analysis.

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。

↑