Reproduce “VAE Rhythm: Temporal Decoding Traces for AI-Generated Video Detection” 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
Ke Sun, Shuwei Li, Haiyang Mei, Mike Zheng Shou
The authors introduce VAE Rhythm, a training-free video-generation detection score that measures periodic changes in reconstruction error as a frozen VAE’s temporal window shifts. The pattern is linked to causal video VAE frame grouping, though it is not universal. On DF26, it reaches 76.2% AUROC, 5.8 points above the best retrained baseline, Effort; this result is benchmark-specific.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.