As in many other computer vision tasks, deep learning has brought revolutionary advances in human behaviour understanding from visual data. Deep models are now extremely effective not only in detecting and analyzing human faces, bodies and collective activities but also in generating realistic human-like behavioral data. From full-body deepfakes to AI-based translation dubbing, deep networks can now synthesize images and videos of humans such as they are virtually indistinguishable from real ones. The workshop will focus on recent advances and novel methodologies for generating human behaviour data, with special emphasis on approaches for forging images and videos depicting real-looking human faces and/or full bodies and on algorithms for detecting fake human-like visual data.
The HBU workshops, organized since 2010 as satellite to ICPR’10, AMI’11, IROS’12, ACM Multimedia’13, ECCV’14 and UBICOMP’15, ACM Multimedia’16, FG’18, ECCV’18 Conferences, aim to inspect developments in areas where smarter computers that can sense human behavior. These events have a unique aspect of fostering cross-pollination of different disciplines, bringing together researchers of mobile and ubiquitous computing, computer vision, multimedia, robotics, HCI, artificial intelligence, pattern recognition, interaction design, ambient intelligence, and psychology. The diversity of human behavior, the richness of multi-modal data that arises from its analysis, and the multitude of applications that demand rapid progress in this area ensure that the HBU Workshops provide a timely and relevant discussion and dissemination platform.
Each edition of the HBU workshop had a different focus theme, dealing with a newly emerging topic or question in the automatic analysis of human behavior. The focus theme of this year is of high interest for computer vision researchers: Generating, Forging and Detecting Fake Human Behavioral Data. The automatic generation of visual contents is currently a very hot topic in the community. With this edition of the HBU workshops, we attempt to foster research on how to generate visual data (still images and videos) describing human behavior both from the applicative and methodological points of view.
