Recent advances in 2D perception have led to very successful systems, able to estimate the 2D pose of humans with impressive robustness. However, our interactions with the world are fundamentally 3D, so to be able to understand, explain and predict these interactions, it is crucial to reconstruct people in 3D. In this talk, I will present our efforts in perceiving 3D humans from images, focusing on increasing the level of accuracy and detail of the recovered representation. I will discuss the challenges associated with this goal, particularly the limited ground truth 3D data, and propose weaker forms of supervision, that are effective in overcoming this limitation.
Georgios Pavlakosis a PhD student in Computer and Information Science, University of Pennsylvania, where he works with his advisor, Kostas Daniilidis. He received the BS degree in Electrical and Computer Engineering from the National Technical University of Athens, in 2014. He has spent time with the group of Michael J. Black at the Max Planck Institute for Intelligent Systems, in Tübingen, and with the group of Yaser Sheikh at Facebook Reality Labs in Pittsburgh. His research interests lie at the intersection of computer vision and machine learning and include reconstruction and pose estimation of objects and humans from single images.
Sponsored in part by Facebook Virtual Reality Lab Pittsburgh