Reproduce “Learning Reconstruction Errors as Spatiotemporal Evidence 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

Jiahua Jiang, Jinglai Li, Jianru Zhang, Jin Zhu

The method detects generated videos by training a video autoencoder and classifier together, then classifying reconstruction errors with 3D convolutions that capture spatial patterns and changes across frames. On HRFVideo, it achieves an average video-level AUC of 94.36 ± 0.12 across five generator subsets, versus 90.71 for DeMamba; results vary by generator and do not provide frame-level localization.

Summary by alphaXiv.

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