Assoc. Prof. Iasonas Kokkinos, Dept. of Computer Science of University College London, to speak at ECE-NTUA on June 6, 2019 at 13:00 (Amphitheater 2)
Talk title: Learning to Align Images and Surfaces for 3D Human Understanding using Deep Learning
Abstract: This talk will present recent work on combining 2D and 3D deformable models with deep learning for the goal of 3D human understanding. Starting from DenseReg and DensePose, two recently introduced deep learning methods for image-to-surface alignment, we will present HoloPose, a real-time method for monocular 3D human pose estimation in-the-wild that combines bottom-up, CNN-based methods for image understanding, with top-down parametric model fitting. Moving on to unsupervised methods for 3D understanding, we will start from Deforming AutoEncoders, a method for unsupervised dense correspondence estimation, and turn to Lifting AutoEncoders, a novel method for unsupervised 3D reconstruction from an unstructured photo collection. We will show that one can disentangle illumination, texture, 3D pose, identity and expression, leading to highly controllable 3D generative models for photorealistic image synthesis.
Bio: Iasonas Kokkinos is Associate Professor in the Department of Computer Science of University College London (UCL) and founder and CEO of Ariel AI. Prior to that he was a Research Scientist at Facebook AI Research (FAIR), a faculty member and researcher at CentraleSupelec and INRIA, and post-doctoral scholar at UCLA. He obtained his Diploma Eng and PhD degrees from NTUA (ECE), and his Habilitation à Diriger des Recherches from Université Paris Est.
His research interests are at the intersection of computer vision and deep learning, aiming at the development of models that unify problems of structured prediction and shape modeling with deep learning, as well as multi-task learning. He publishes and reviews regularly in the major computer vision conferences (CVPR, ICCV, ECCV), has served multiple times as Area Chair and has been an Associate Editor for CVIU and IVC
The event is organized by the "Signals, Control and Robotics" Division.