Reproduce “No-Reference Rendered Video Quality Assessment: Dataset and Metrics” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sipeng Yang, Jiayu Ji, Qingchuan Zhu, Zhiyao Yang, Xiaogang Jin
Researchers from Zhejiang University and OPPO Nanjing Research Center created ReVQ-2k, the first large-scale no-reference video quality assessment (NR-VQA) dataset specifically for rendered videos, which includes novel temporal stability annotations. Their proposed two-stream NR-VQA metric, designed to assess both image quality and rendering-specific temporal stability, achieved an average 3.9% PLCC and 3.2% SRCC improvement over state-of-the-art baselines on ReVQ-2k. The metric successfully evaluated closed-source mobile supersampling and frame generation strategies in real-time rendering scenarios.
Key insights
- Rendered videos exhibit distinct temporal artifacts, such as flickering and moving jaggies, which are critical to user experience but often overlooked by NR-VQA methods designed for camera-captured content.
- Explicitly evaluating temporal stability alongside traditional image quality factors is essential for accurate and perceptually aligned quality assessment of rendered videos.
- Subjective annotations for temporal stability (TS-MOS) provide valuable supervision for training models to effectively identify and quantify rendering-specific dynamic degradations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.