Face Recognition with Synthetic Data
摘要
In the last few years, face recognition has achieved extraordinary progress in a wide range of challenging problems including pose-robust face recognition [5, 24, 63], matching faces across ages [15, 17, 56, 60], across modalities [13, 14, 16, 30, 31], and occlusions [40, 49, 71]. Among these progresses, not only the very deep neural networks [22, 25, 29, 48] and sophisticated design of loss functions [10, 23, 32, 57, 61], but also large-scale training datasets [20, 26, 27] play important roles.