Our research interests focus on: (1) what are the ground truth representations of image and video, and how to modeling and learning them from natural images and videos based on small set of samples? (2) How to integrate descriptive models, such as global or local appearance models largely used in the vision literature, and generative models, such as sparse coding, etc. and traditional structure- or grammar- based models, eg. stochastic context free grammar(SCFG)? (3) How to integrate and schedule bottom-up discriminative computing and top-down generative computing in the Bayesian framework, and how to use top-down generative models to help bottom-up discriminative model compute more effectively?

At present, Lotus hill institute(LHI) sets up the following five labs.

Lab I : Fundamental Research in computer and human vision

(Please click each button to see detail information)

Lab II: Lotus hill institute Image and Video Database for the ground truth of vision

Research Directions and Goals:

• Study the ground truth representations of images and videos based on stochastic context sensitive grammar embeded in And-Or graph representaion.
• Build up a large scale human annotated image and video database for the ground truth of modeling and learing in vision.
• Build up international benchmarks with different levels for all the levels in vision, such as segmentation, detection, recognition, etc.
• Gradually set up Lotus hill institute's standard of image and video database.

Lab III : Large Scale and General Image Parsing System

Research Directions and Goals :

• Researches on the basic and core problems in vision and image science under a unified representational and computing framework.
• Study inferential algorithm for general image parsing problems.
• Study the compositionality on representation and computing, and build up large scale recursive inferential framework.

Lab IV : Video Parsing

Research Directions and Goals :

• Researches on complex motion and event parsing and synthesis theory based on event grammar.
• Video repairing, composing and stylization.
• Motion and dynamics analysis, information retrieval.

Lab V : Digital Arts

Science and art always maintain subtle relationships between each other. It is said that "science assures, art upsets". However, while scientists' acute insights and precise thoughts and artists' great creativeness and expressive forces are often thought to be opposed, they really share some common details. In fact, we human's world is a discrete and limited one, including science and art, in which treasures are of selection. We would like to discover in a scientific way, the principles according to which people choose their favorite artworks, and furthermore, tell people in a scientific way too, WHAT IS ART.