Niloy J. Mitra

The 3D world around us is complex and often cluttered. We, as humans, find it second nature to easily reason about our surrounding, understand the function and purpose of most objects, and navigate the world accordingly. Performing the same computationally remains challenging. Capturing and representing object geometry in a compact, yet expressive, digital form is not only an essential component for most modeling and simulation systems, but is also fundamental to learning natural laws and forms. Mainly I focus on,

(i) Shape Analysis: Humans possess superior reasoning and observation skills and are considered to be experts in many areas like bio-engineering, medicine, etc. However, with the huge amount of digital data now being easily available, automation is desirable. A common theme is to understand important relations between objects and object parts in 3D scenes, e.g., identifying symmetries, i.e., equivalence of object parts under certain transformations, as well as to detect regular structures in 3D geometry. In recent efforts, we have started exploring the power of shape abstraction, high level object understanding, and its application to context-aware deformations for geometry processing and shape visualization.

(ii) Exploring and Exploiting Data Collections: An intriguing and challenging topic is to identify and encode model variations across collections of similar datasets using statistical techniques. This problem is specially relevant with large volumes of 3D geometry quickly becoming available via online repositories, while we still lack effective tools for browsing and navigating such datasets, e.g., understanding geometric optimization and manufacturing constraints to better navigate the space of constrained meshes, but still respecting both local and global constraints. This line of research opens up interesting directions in fabrication-aware design and exploration, and optimizing directly on such design manifolds.

For details, please visit my PUBLICATIONS page.