Colloquium Cognitive Systems

Visual inference in human and artificial systems

Prof. Tom Wallis, Ph.D. (TU Darmstadt)

Abstract:
Humans understand visual scenes from just a glance or two: we effortlessly perceive the structure of our environment, knowing which parts of the retinal input belong together and which do not. Despite incredible progress in artificial seeing, these systems still fall far behind our perceptual systems in terms of breadth and robustness in these segmentation processes. I will present work from my group in which we seek to understand how humans use visual features to perform search and segmentation, using a cue combination approach. We are also developing a rendering pipeline to study this problem systematically in artificial systems. Finally, I will present work using differentiable rendering to probe artificial visual systems for 3D scene information.
About:
Thomas Wallis is an experimental psychologist who uses behaviour, eyetracking and computational modelling to study visual perception in humans and machines. He has done postdoctoral fellowships at Harvard Medical School and The University of Tübingen. Since 2021 he leads the AG Perception in the Centre for Cognitive Science at the Technische Universität Darmstadt.