Lab Home | Phone | Search | ||||||||
|
||||||||
3D Deep Learning for Shapes is a heated and active field of research in computer science, trailing the immense success achieved in the fields of 2D imaged based computer vision algorithms. 3D Deep Learning also finds much more direct applications in physical and engineering systems, where most problems are 3D in nature. However 3D learning is inherently different from its 2D counterpart in two important ways: its O(n3) growth in memory consumption, and the sparse nature of shapes. Hence 3D learning still lacks the robustness to be used towards engineering problems. In this talk, three projects will be discussed, illustrating different aspects of shape learning: shape generation and shape learning for inference of physical properties, and last but not least our ongoing work improving 3D shape learning. Host: Balu Nadiga |