Reproduce “Generative Action Tell-Tales: Assessing Human Motion in Synthesized Videos” 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
Xavier Thomas, Youngsun Lim, Ananya Srinivasan, Audrey Zheng, Deepti Ghadiyaram
The authors introduce two human-motion video scores: Action Consistency compares a clip with real examples of its intended action, while Temporal Coherence measures changes in frame embeddings. Across 300 videos and ten actions, the scores correlate with human ratings at 0.61 and 0.64, versus GPT-5’s 0.45 and 0.38. External testing supports only temporal ranking.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.