CS Forum: Victor Lempitsky, Samsung AI Center & Skoltech "Drawing Humans with Neural Networks"
Drawing Humans with Neural Networks
Samsung AI Center & Skoltech, Moscow
I will present two recent works from Samsung AI Center on realistic rendering of humans with deep convolutional neural networks. The first work develops an approach that creates photorealistic personalized head avatars from few images of a person (few-shot avatars). The resulting avatars can be driven by facial keypoints. To be able to capture a person's appearance from few images, this system uses a strong prior learned from a large and diverse dataset of human videos of different people. The second work develops a system for capturing the static appearance of a single person in 3D (a 3D portrait). The new system is in many ways similar to standard 3D modeling framework, however it uses a point cloud geometric representation together with a deep neural rendering network. Such neural point-based graphics system builds a 3D model that can be rendered with photorealistic quality from new viewpoints, without performing the brittle step of meshing / surface estimation.
Victor Lempitsky leads the Samsung AI Center in Moscow as well as the Vision, Learning, Telepresence (VIOLET) Lab at this center. He is also an associate professor at Skolkovo Institute of Science and Technology (Skoltech). In the past, Victor was a researcher at Yandex, at the Visual Geometry Group (VGG) of Oxford University, and at the Computer Vision group of Microsoft Research Cambridge. He has a PhD ("kandidat nauk") degree from Moscow State University (2007). Victor's research interests are in various aspects of computer vision and deep learning, in particular, generative deep learning. He has published extensively in top computer vision and machine learning venues and has served as an area chair for top computer vision conferences (CVPR, ICCV, ECCV, ICLR) on multiple occasions. His recent work on neural head avatars was rated as the most-discussed research publication of 2019 by Altmetric Top 100 rating.
Professor Juho Kannala, Department of Computer Science