Towards a fully automatic solution for face occlusion detection and completion

Date: Thu 15 Oct 2020
Time: 09:00 AM
Venue: http://meet.google.com/xqj-mswe-wbv
Speaker: Iacopi Masi

Title: Towards a fully automatic solution for face occlusion detection and completion

Abstract: Computer vision is arguably the most rapidly evolving topic in computer science, undergoing drastic and exciting changes. A primary goal is teaching machines how to understand and model humans from visual information. The main thread of my research is giving machines the capability to (1) build an internal representation of humans, as seen from a camera in uncooperative environments, that is highly discriminative for identity (e.g., person re-identification and face recognition); and (2) to semantically analyze human faces to detect, segment, reconstruct, and synthesize them (e.g., occlusion detection and face completion). In this talk, I show how to enforce smoothness in a deep neural network for better, structured face occlusion detection and how this occlusion detection can ease the learning of the face completion task. Finally, I quickly introduce my recent work on Deepfake Detection.  

Bio: Dr. Iacopo Masi is a Research Assistant Professor in the Computer Science Department of the Viterbi School of Engineering at the University of Southern California (USC). He is also a Research Computer Scientist at the USC Information Sciences Institute (ISI). Dr. Masi earned his Ph.D. degree in Computer Engineering at the University of Firenze, Italy. Immediately after, he moved to California and joined USC, where he was a postdoctoral scholar. Dr. Masi has been Area-Chair of several WACVs and currently serves as Associate Editor for The Visual Computer - International Journal of Computer Graphics. He organized an International Workshop on Human Identification at ICCV'17 and was Workshop Chair at SIBGRAPI'18. Dr. Masi's main research interest lies in solving the computer vision problem, specifically, the subjects of tracking, person re-identification, 2D/3D face recognition, and modeling.

© Università degli Studi di Roma "La Sapienza" - Piazzale Aldo Moro 5, 00185 Roma