Fall 2017 GRASP Seminar Series: Shubham Tulsiani, UC Berkeley, "Consistency, Commonality and Compositionality : Learning 3D Structure without 3D Supervision"

ABSTRACT

In this talk, I will discuss the task of learning to infer 3D structure without explicit supervision and present two recent attempts in this direction. I will first describe a differentiable ray consistency formulation which enables learning single-view 3D prediction models using indirect multi-view supervision. We will show that this formulation allows leveraging varying kinds of observations (foreground labels, depth or semantics) as supervisory signal and examine its application in diverse scenarios. I will then present a method that learns to assemble shapes using volumetric primitives and show that this yields interpretable and coherent abstractions in an unsupervised manner. I will demonstrate that these representations can be leveraged for applications like shape parsing, manipulation, retrieval etc.

Presenter's biography

Shubham Tulsiani is a graduate student in the Computer Vision group at UC, Berkeley where he is advised by Prof. Jitendra Malik. His research interests lie at the intersection of recognition, pose estimation and 3D reconstruction from a single image. His work is supported by a Berkeley Graduate Fellowship.