Reproduce “Hard Sample-Conditioned Decorrelated Attention for Robust Face Recognition” 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

Tahar Chettaoui, Guray Ozgur, Eduarda Caldeira, Naser Damer, Fadi Boutros

ViT-HarDis adapts attention-head decorrelation during face-recognition training to identity ambiguity. It weights the penalty more for correctly classified samples close to a competing identity center, without changing the inference model. On a controlled IJB-C comparison, it reaches 80.52% TAR at FAR 10^-5 versus 78.51% for uniform disentanglement, but scores below baseline on IJB-B at that threshold.

Summary by alphaXiv.

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