Reproduce “TwinD: Generalizing Deepfake Detection from Blending to Entire Face Synthesis” 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

Andrii Yermakov, Jan Cech, Mario Fritz, Jiri Matas

TwinD trains a facial deepfake detector on FaceForensics++ fakes and artificial twins made from real frames. Reconstruction twins combine autoencoder output with restored real pixels; blob twins blend altered frames with originals. Paired training targets decoder traces and blending artifacts. Across 19 datasets, it reaches 96.3% mean video-level AUROC, including 95.6% on EFS, but reports greater sensitivity to noise and downscaling.

Summary by alphaXiv.

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