Multi-supervised metric learning for fisher vector faces

Metric learning has been widely used in face verification. However, most existing metric learning methods only have one single supervised goal, which is insufficient. This paper makes two contributions: first, we show that the multi-supervised metric learning on Fisher vector faces is better than the original one, and is capable of outperforming the state-of-the-art face verification performance on the challenging “LFW” benchmark on condition of 2D-alignment. Second, we show that patch-based alignment and 3D-alignment is useful to Fisher vector faces, and can improve the final result.