Image translation between human face and wayang orang using U-GAT-IT

Abstract

The people's puppets (wayang orang) performance typically requires approximately one hour for the performers to assume the role of a wayang orang, as this duration is necessary to apply makeup and select suitable attire. One potential solution to this issue involves the creation of a computerized simulation that replicates the process of putting makeup and traditional clothing on the face and head of the wayang orang performer. The completion of this work can be achieved through the utilization of image translation techniques. The objective of this study is to employ the unsupervised generative attentional networks with adaptive layer-instance normalization for image-to-image translation (U-GAT-IT) technique to convert human faces into wayang orang representations. The study utilizes an unpaired dataset comprising 1216 training data samples and 240 testing data samples. The primary objective of this study is to effectively preserve both the background picture and the facial identification component inside the given input image. This study utilizes quantitative assessment methods, specifically kernel inception distance (KID), Frèchet inception distance (FID), and inception score (IS), to evaluate the quality of the generated output image produced by the generator. Experimental results demonstrated that U-GAT-IT outperforms dual contrastive learning generative adversarial network (DCLGAN) in terms of the metrics IS, FID, and KID.