Video quality assessment improved with artifact-based reward modeling
ArtiVidRM: Verifiable Artifact-as-Rubric Reward Modeling for Explicit Quality Articulation in Video Generation
Computer Vision and Pattern Recognition
Summary
Video generation can be improved by teaching computers to judge video quality, but existing methods often give vague scores without clear reasons. The authors propose a method called ArtiVidRM that first finds visible errors in the video and lists them out. These lists act like rubrics explaining why a video gets a certain score, making the judgments clearer and more useful for improving video generation. Their approach also trains the system to better predict these error lists and rank videos, resulting in better video quality assessment and helping improve video generators.
What this means in practice
- •For video generation developers: Improve video generation systems by incorporating ArtiVidRM’s artifact-based reward signals for more accurate quality evaluation and enhancement.
- •For content moderation teams: Use explicit artifact detection to automatically identify and flag quality issues in user-generated videos during platform content review.
Authors
Caorui Li, Yangguang Ji, Shuxia Lin, Yu Chen, YiCheng Xiao, Yongliang Wu, Zhenxiang Jiang, Licheng Tang, Deng Li, Zhuxuanzi Wang, Jiahao Sun, Chonghan Liu, Lirian Su, Jay Wu, Xu Yang
Abstract
Reinforcement learning has become an effective approach for improving video generation, but its success depends heavily on reliable reward models. Existing methods either map videos directly to opaque scalar scores or generate free-form rationales that are not explicitly optimized as the basis of the final judgment. We propose ArtiVidRM, a verifiable, artifact-as-rubrics video reward model that follows an observe-then-score paradigm. It first identifies visible generation artifacts and organizes them into a structured artifact list, then uses these artifacts as explicit rubrics for quality assessment. This intermediate representation grounds each score in concrete video-level evidence, making the reward more interpretable and informative for optimization. We further introduce a multi-reward reinforcement learning strategy to jointly optimize artifact prediction and quality assessment. The reward decomposes into complementary objectives that evaluate artifact prediction quality, including semantic coverage and prediction efficiency, together with a pairwise preference objective that supervises quality ranking through score differences. By jointly optimizing the intermediate artifact representation and the downstream preference prediction, the model learns to produce informative artifact lists while improving the accuracy of video quality assessment. Experiments show that ArtiVidRM achieves state-of-the-art performance on video preference benchmarks, provides effective signals for video diffusion optimization, and enables artifact-guided refinement of closed-source video generators.