GaitFace: A Multimodal Dataset for Long-Range Person Identification
Date:
Fifteen-minute oral presentation in the main conference.
GaitFace is a public dataset for long-range person identification at border crossings. It contains face and gait data from 70 consenting participants across two sessions, about 2.7 TB in total, captured outdoors at distances of up to 100 m from a ground-level and an elevated (~9–10 m) viewpoint. It pairs smartphone pre-enrollment data with in-the-wild captures from multiple cameras and angles. Benchmarks of state-of-the-art face and gait models show that they fail at low resolution and elevated viewpoints without optical zoom: at 100 m the average face is only 18×22 pixels.
Authors: Alain Komaty, Luis S. Luevano, Vidit Vidit, Anjith George, Zeina Al Amine, Sébastien Marcel.