Deep Learning for 3D Recognition

The goal of this project is to leverage deep learning to explore new methods for 3D recognition. A large focus will be on finding object instances in images, along with their semantic as well as geometrical properties (such as 3D shape and pose). Starting out with known rigid objects, the goal is to consider more general scenarios as well. A recurring theme will be how to gain most from the combination of learning, white-box modeling and geometry.