Reproduce “Content-Rich AIGC Video Quality Assessment via Intricate Text Alignment and Motion-Aware Consistency” 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
Shangkun Sun, Xiaoyu Liang, Bowen Qu, Wei Gao
CRAVE, developed at Peking University, is a multi-dimensional video quality assessment framework tailored for content-rich, next-generation AI-generated videos, featuring intricate text alignment and motion-aware consistency. The accompanying CRAVE-DB benchmark provides 1,228 videos from state-of-the-art models with detailed human annotations, demonstrating leading performance on CRAVE-DB with an SRCC of 0.7587.
Key insights
- Integrating textual semantics at multiple granularities (paragraph, phrase, word) significantly improves the precision of text-video alignment in content-rich AIGC videos.
- A hierarchical approach that combines both low-level optical flow and high-level action recognition effectively identifies and quantifies nuanced, AIGC-specific motion distortions.
- Creating dedicated benchmarks with videos from state-of-the-art T2V models and complex, long-form text prompts is essential for developing VQA models capable of evaluating and generalizing to future high-quality AIGC content.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.