compression

FaceMap — perceptual facial saliency maps driven by visual distortion, for better avatar faces. Learned from 45+ participants pairwise study, SROCC 0.82.