This is a vision / graphics seminar based in the Ithaca campus. It is predominantly a platform for PhD students and researchers to present their ongoing or unfinished work and solicit feedback, although occassionally we might have external speakers. Students interested in vision / graphics research are invited to attend.
The seminar takes place on Mondays, 4-5 pm in Gates 310. This semester, the first seminar will be on September 17.
There is a mailing list associated with this seminar : gvsem-l. Email bharathh-AT-cs-DOT-cornell-DOT-edu if you cannot join.
Abstract: A broad range of robotics and mixed reality systems are built on top of traditional Simultaneous Localization and Mapping (SLAM) algorithms. This talk will discuss the frontend/backend distinction in SLAM as well as discuss our group’s recent research on ConvNet-based frontends for Visual SLAM. The talk will focus on Deep Learning architectures for geometric computer vision tasks as well as novel self-supervised methods for training such networks when ground-truth training data is difficult/impossible to obtain.
Bio: Tomasz is a Principal Engineer at Magic Leap, focusing on applied research at the intersection of Deep Learning, SLAM, and Mixed Reality. He obtained his PhD from Carnegie Mellon’s Robotics Institute in 2011 under the supervision of Alexei A. Efros, spent two years as a postdoctoral researcher at MIT CSAIL, and co-founded a small computer vision startup in Boston.Talk Keywords: Perception for Mixed Reality, Visual SLAM, Descriptor Learning, Self-Supervised Learning, Fully-convolutional ConvNets, Geometric Deep Learning Relevant Links:
The calendar is below.