Research Topics for Perspective MS and PhD Students
1) Shape Understanding
2) Sketch-based Scene Composition
3) 3D Scene Labeling
4) Facial Performance Capture
5) 3D Printing
6) Virtual Medicine


Shape Understanding

Motivation: Ease of 3D data acquisition (Kinect, smartphones, depth sensors, etc.) brings up a growing need for automatic algorithms that analyze the semantics and functionality of digitized shapes. Analysis algorithms, such as shape segmentation, feature detection, symmetry analysis, and surface correspondence, are employed to search, organize, synthesize, and edit 3D shapes.

Apps: Video games, virtual shopping, medical apps, industrial simulations, documentation of cultural artifacts, etc.

Conventional approach: Analyze based on geometric or topological features computed directly from the shape itself.
Innovative approach: Analyze based on how external agents interact with the shape.

Conventional approach: Shape correspondence between a pair of shapes in isolation.
Innovative approach: Shape correspondence over multiple shapes at once using contextual info.

Conventional approach: Shape interpolation on geometric level.
Innovative approach: Shape interpolation on geometric and topologic level.

Sketch-based Scene Composition

Motivation: Given the high interest to the video games and animation movies, we can safely claim that there are many people enjoying the virtual 3D world. Most of them, however, are in consumer position. By providing them an intuitive and easy-to-use 3D modelling system, we can help them be producers and reveal their creativity.

Apps: Video games, movies, education, etc.

Conventional approach: Match 3D shapes with 2D sketches by projecting the shapes to 2D.
Innovative approach: Match 3D shapes with 2D sketches by implicitly lifting the sketches to 3D.

3D Scene Labeling

Motivation: We want a robot to execute commands like "Fetch me the mug on the table". This requires the labeling of the raw RGB-D scene data coming from affordable scanners such as Kinect.

Apps: Robotics, games, movies, etc.

Conventional approach: Analyze scene by recognizing one object at a time.
Innovative approach: Analyze scene by exploiting the interactions between the constituent objects.

Facial Performance Capture

Motivation: Transfer yourself along with your mimics and movements to the other side in real-time, a different step towards holographic communication.

Apps: Video games, enhanced instant messaging, education, etc.

Conventional approach: Register a template to the user's face using prior knowledge.
Innovative approach: Register a template to the user's face without prior knowledge.

3D Printing

Motivation: Current 3D printer technology imposes specific constraints on digital shapes to be fabricated, e.g., size, geometry, etc. Besides, external forces such as gravity need to be simulated on the digital elastic models before printing them out to the real world. Support material optimization is also of interest.

Apps: Industry, architecture, medical apps, etc.

Conventional approach: n/a as this is a relatively new field.
Innovative approach: think.

Virtual Medicine

Motivation: Exploit the dense data in medical images to the fullest and quantify medical conditions objectively in a systematic way. Give doctors a virtual practicing environment before dealing with real patients. Infer the inner body from the outer surface (X-rays, scanners), and vice versa, to decrease or eliminate unhealthy radiation. Show patient how she would look like before aesthetic surgery.

Apps: Medical applications.

Conventional approach: Register dense medical images directly to each other.
Innovative approach: Register dense medical images through a compact intermediate space.

Back to the home page