We developed a physics-based particle method to simulate various fluid mixture effects. We introduced diffusion and chemical reaction to model them. Written in C++ and OpenGL. See details.

Based on SFS and image understanding techniques (e.g. intrinsic image), we focused on optimizing solutions for lighting synthesis based on a single color image with minimal user-interventions. Written in MATLAB but incomplete.

This is the course project for CSC2521, Artificial Life. Based on the popular flock model, Boids, message passing and dispatching are introduced to improve coordinated group behaviors in complex environments. Written in C++ and OpenGL.

This is the course project for ECE1772, Motion Analysis in Computer Vision. The project was solely MATLAB implementation of ICCV99 Multi-frame Optical Flow Estimation Using Subspace Constraints. Written in MATLAB.

This is the course project for CSC2515, Machine Learning. Principal curves are smooth curves that minimizes the average squared orthogonal distance to each point in the dataset. Two different methods for finding principal curves were chosen from the literature and implemented to show and compare their estimation performances. Written in C++.

This is the course project for CSC2522, Advanced Image Synthesis. Most work involved the implementation of SIGGRAPH2001 Illustrating Smooth Surfaces but some concepts (e.g. cusp detection and smoothing) are replaced by simple methods. For example, in order to smooth hatching strokes, I used a hierarchical clustering information obtained from mesh adaptations. Written in C++ and OpenGL.

We have developed an efficient method for interactive multi-resolution video that allows for varying spatial and temporal resolutions in any part of a video sequence. While conventional interactive video has mainly concentrated on interactive story telling or scenario, out representation allows the viewer to interactively control the spatial and temporal resolution at real-time. Written in C++ and MFC.

We observe that A* algorithm does not perform very well for finding optimal paths in the dynamic terrain map that varies constantly. We introduced following two concepts: layered critical regions and hierarchical approach. They are shown to handle abrupt changes of the map with efficiency and less overhead. Written in C++ and MFC.