Reproduce “FVScore: Specification-Driven Neuro-Symbolic Video Evaluation” 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

Do Xuan Long, Long T. Le, Yiwen Song, Min-Yen Kan, Tomas Pfister, Yale Song

FVScore evaluates generated videos by translating prompts and observed events into explicit and inferred claims, then checking them against visual, audio, and cross-modal evidence. It uses soft temporal operators to produce scores and timestamped diagnoses. On FVBench, its video–audio synchronization Pearson correlation with human ratings is 0.678, versus 0.239 for VISTA’s MMAC; this is a complex-benchmark result, not a universal ranking.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.