Ryota Natsume*, Kazuki Inoue*, Yoshihiro Fukuhara, Shintaro Yamamoto, Shigeo Morishima, Hirokatsu Kataoka (*equal contribution)

Understanding Fake-Faces

Brain-Driven Computer Vision Workshop (BDCV 2018) in ECCV 2018, Posters

http://www.upcv.upatras.gr/BDCV/index.php

The research of face is one of the most active topics in computer vision, and deep neural networks (DNNs) fill the gap between human-level and computer-driven performance in face verification. The performance gap seems to get close in the accuracy-based expectation.  However, a curious scenario appears, namely face understanding is really close to the human-level performance? In the paper, to confirm the brain-driven concept, we conduct image-based detection, classification and generation on our Fake-Face database. The Fake-Face database has two configurations: (i) false positives by face detector (e.g. VJ, CNNs), and (ii) Simulacra faces which seem to be human faces but there is no real-face. The result shows a suggestive knowledge, such as the recent vision-based face understanding still remains the gap from the human-level performance. Positively, we obtain the knowledge to improve the current face understanding model toward the next level.