DriveFace: A Cross-Spectral Through-Glass Face Dataset for On-the-Move Vehicular Border Control
Date:
Two-minute poster spotlight in the main conference, followed by the poster session.
DriveFace is a dataset for on-the-move face recognition at vehicular border control, where travellers are identified inside their cars. It pairs visible-light smartphone pre-enrollment with near-infrared (NIR) captures through automotive glass, for 70 subjects over two sessions, five window tint levels, and ages 18–85. It also includes DriveFace-PAD, a subset of print, replay, and mask attacks. In our baselines, face recognition reaches 96–97% Rank-1 despite the RGB–NIR gap, but tinted glass remains the hardest setting. Attack detection works well for known attacks (0.5% ACER) but not for unseen ones (up to 45.8% ACER).
Authors: Anjith George, Luis S. Luevano, Alain Komaty, Zeina Al Amine, Vidit Vidit, Sébastien Marcel.
Spotlight slides
Poster
Paper page
Project page
Code
Dataset