GBU-Palm: A Multimodal Video Dataset and Benchmark for Palm Presentation Attack Detection
GBU-Palm pairs 21,326 palm videos with leakage-controlled tests for spoof detection across environments.
The dataset covers 105 subjects, 210 palms, six acquisition settings, and bona fide, print, and replay presentations. It includes 6,310 synchronized RGB-NIR samples. The benchmark finds models degrade differently under environmental shift, and RGB-NIR fusion does not reliably beat RGB alone. The authors also break down failures by accept/reject outcomes and probe how architectures use spectral and temporal evidence. ArXiv · AI/CL/LG's note
The dataset covers 105 subjects, 210 palms, six acquisition settings, and bona fide, print, and replay presentations. It includes 6,310 synchronized RGB-NIR samples. The benchmark finds models degrade differently under environmental shift, and RGB-NIR fusion does not reliably beat RGB alone. The authors also break down failures by accept/reject outcomes and probe how architectures use spectral and temporal evidence. ArXiv · AI/CL/LG's note
score 4