Reproduce “Masked Face Recognition: Human vs. Machine” 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
Naser Damer, Fadi Boutros, Marius Süßmilch, Meiling Fang, Florian Kirchbuchner, Arjan Kuijper
The study by Damer et al. comprehensively compares the performance of human experts and automatic face recognition systems in verifying masked faces, revealing significant performance degradation across all systems and highlighting the limitations of current synthetic mask generation methods. Human experts with wider contextual views generally outperformed tighter crops, and top-performing automatic systems experienced EER increases from 0% to 1.0-2.8% when facing real masked probes.
Key insights
- Face masks induce significant performance degradation in both human and automatic face recognition systems, primarily by shifting genuine similarity scores towards imposter scores.
- Human experts benefit from wider face crops, utilizing peripheral information (like hairstyle), and can sometimes outperform top-performing automatic systems under masked conditions.
- Current synthetic mask generation methods do not adequately replicate the challenges of real masks, leading to an overestimation of performance when used in evaluations for both humans and machines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.