Learning 3D Perception from Single Images - Unifying Learning and Geometry
Explore a colloquium presentation on "Learning 3D Perception" by Shubham Tulsiani from Facebook AI Research. Delve into the integration of learning-based techniques and geometry-driven methods for single image 3D understanding. Discover how incorporating the concept of 2D percepts as projections of a 3D world can bypass direct supervision needs. Learn about leveraging the principle that complex 3D structures have simpler components to infer factored 3D representations. Gain insights from Tulsiani's research, which aims to enable machine perception systems to learn about and infer the structure underlying the physical world. Understand the significance of this work, which was awarded the 'Best Student Paper Award' at CVPR 2015. This 52-minute presentation, recorded on February 4, 2020, includes closed captions and is part of the Paul G. Allen School colloquium series.