Architectural Bias in Face Presentation Attack Detection: A Comparative Study of Vision Transformers and Convolutional Neural Networks
本研究针对人脸活体检测（Face Presentation Attack Detection, PAD）系统中存在的种族偏倚问题，系统比较了Vision Transformer（ViT）架构与传统卷积神经网络（CNN）在跨种族场景下的表现。