Reproduce “Foundation Models for Face Presentation Attack Detection: A Unified Linear-Probing Benchmark” 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
Peter Lorenz, Anjith George, Sébastien Marcel
The authors evaluate 24 frozen vision encoders for face presentation-attack detection using trained linear classifiers. The probes test whether pretrained representations separate bonafide faces from attacks across four MCIO datasets under single-source training. InternViT-6B has the lowest mean intra-dataset ACER, 1.6%, while CLIP ViT-B/32 has the lowest mean cross-dataset ACER, 31.6%. These frozen probes show cues, but do not reliably transfer across datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.