We are living in the age of large data. Vast amounts of data is collected form the real-world by various sensors or simulated to produce images, videos, depth images, volume images, 3D geometry, and other forms representing the geometric structure of the world. Our main goal in this project is to develop novel mathematical and computational techniques and tools that can analyze, reconstruct, and process general complex geometric structures represented by acquired or simulated data.
In the last decades, many algorithms that can process geometric data have been proposed in the point-based computer graphics literature. Most of these methods work under the assumption that the data resides on a smooth manifold. However, as the data becomes more accessible, it also gets more complex. This renders the current assumptions invalid. To deal with complex data, it is essential to drop the common smoothness and manifoldness assumptions and develop a more general framework for non-manifold data.
Body Shape Reconstruction
The goal is to investigate novel methods in geometry synthesis and texture synthesis to tackle the problem of reconstructing a body shape. Data-driven approaches based on body poses and silhouettes are used to complete human shapes and textures, in order to plausibly regenerate hidden parts of the scene with desired resolution.
Image and Video Processing
Images and videos are everywhere and represent a huge amount of visual data. Our goal is to develop geometry-based image and video processing methods, in order to understand and utilize them.