Recently we are doing mesh morphing based on our spherical parameterization. By constructing spherical meshes for zero-genus meshes, feature alignment and merging two meshes become much easier. I am trying to develope something like "spectra morphing" to control the morphing process. The left video is the morphing process from  Mannequin to Venus head, with 18 feature points marked.

We proposed a novel framework for digital signal processing over arbitrary meshes. All mesh attributes are transformed into spherical signals through our robust spherical parameterization algorithm. Harmonic analysis and fast harmonic transform, the powerful analysis tools on unit sphere,  are empolyed to analysis these spherical signals. Within our framework, traditional signal processing operations such as filtering, enhancement and convolution can be extended to 3D geometry naturally. The left video demonstrates the filtering process of the Feline model. The Feline becomes smoother and smoother and finally it turns to ellipsoid. This is analogous to low-pass filtering an image with a huge Gaussian filter. For more demos, click here.

By parameterizing a mesh onto a square, all mesh attributes are transformed into 2D signals. JPEG2000 standard for still images is employed to compress these 2D signals. Compared with previous geometry compression schemes, the superiority of our geometric signal compression algorithm includes the following aspects: consistent compression of all mesh attributes, a better tradeoff between bit rate and distortion and easy integration into existing image compression standards.  The left video demonstrates the progressive recovering of the color information of Stanford Bunny. This is analogous to the progressive transmission process of  still images.

Based on our geometric signal processing framework, we developed a multi-resolution eding system. Users first edit vertex positions, colors and normals at a low resolution mesh. Next the spectrum differences between the modified signal and the signal before modification are calculated and added to the original signal defined on the high resolution mesh. The resulting mesh will look like what users want it to be. Our editing scheme shows the power of our geometric signal processing framework, making researchers freely move back and forth between spatial domain and frequency domain. The left video shows the editing result of Cat model. We  performed pushing, rotating and scaling operations on it.

By parameterizing a mesh onto a square, traditional image watermarking techniques can be extended to 3D models. Since all attributes are transformed into 2D signals, our scheme can add watermarking into any attributes. We have carried out a lot of experiments using several image watermarking algorithms based on DCT and DWT and got high dectection rates even after very hard attacks.

In the summer of 1999, I worked as an intern at Microsoft Research China. One of my research tasks  is to implement concentric mosaic array, using a vertically aligned camera array to capture an array of mosaics. Compared with the well known concentric mosaic, this method can eliminate the vertical distortions and provide significant vertical parallax and lighting changes as well. My another task is to combine synthetic objects and real scenes captured by cameras.

Using Debevec's scheme of modeling architecture from photographs (SIGGRAPH 96), we implemented a image based modeling system. With this system we modeled several buildings in Zhejiang University. 

Many computer graphics applications require complex, highly detailed models to maintain a convincing level of realism. Consequently, models are often created or acquired at millions of polygons, far more than current PCs or workstations can render at interactive frame rates. Level of detail is a key technique for real-time rendering. Our group has carried out a lot of rearch on this direction in the past few years. For more details, you can visit Prof. Jiaoying Shi's homepage. 

It's the first software I developed on PC using Visual C++. We tried to develop our own 3D toolkit offering a solution to interactive graphics programming, just like Open Inventor. My partner, Dr. Rui Lin, got 500K RMB investment from a businessman for this project and founded a software company. However, when we finished it after half a year,  we found no one would buy it. Finally this software won the first prize in the 1998's National College Students' Software Contest. We didn't make any money from it. After our company collapsed, Mr. Rui Lin rewrote the software and renamed it Intra3D. Its source is open. Get it here.

I developed this little game just for learning the "select" function of OpenGL. Later, I added the network and chart  funcition for fun. Click here to get the source code (written in Visual C++).

EasyHome is a CAD system for designing and upholstering houses, like 3D Home. Users first draw  2D drafts for the houses they want, the 3D models of the houses are produced automatically. Next users select and place all kinds of furnitures in the houses as they like. They can also decorate the houses with different wallpapers and carpets. Finally users can walkthrough the houses. This software is developed not for architecture experts but for common people. Our goal is to make the housing easy to people. I developed this system in PioneerSoft Inc.

EasyBowling Game Machine is developed based on virtual reality and computer vision technique. Our purpose is not only to simulate the Bowling Game in computer but also to provide a real bowling game feeling (including visual feeling and physical strength feeling) to users. We employ the real bowl, but our bowling track is only about 3 meters long which is much shorter than the real bowling track (about 20 meters). In order to detect the motion of the bowl, we use a PC Camera to monitor the track, get the position and the velocity of the bowl, then calculate the movement and the collision of the bowl, at last, we render the result to a TV screen in real time, player can see the result immediately after he(or she) throws the bowl. This machine is a commercial product of  PioneerSoft Inc. It ranked 19th in 2000 Young Inventors Awards held by Far Eastern Economic Review.

This is a research software, performing mesh simplification based on edge collapses. I adopt the well known "Quadric Error Metric" by Graland (SIGGRAPH 1997). You can find a better software QSlim written by Garland. In this software I only support our own simple file format ".tm". You can view the tm files directly to check its format. I am developing a quite powerful version for our DGP toolkit. Click here to download the source code (written in Visual C++) and some data are available here.

I founded PioneerSoft Studio in 2000. This studio dimissed after 7 months, since I joined PioneerSoft Inc. During that 7 months, my personal interests focused on network software. I wrote prototypes of server popular protocols, such as POP3, SMTP, HTTP, FTP. You can download my simple server implementations for learning, email server, http server and ftp server. The other 2 members of the studio also developed some interesting stuffs, such as Easy Camera (Mr. Jin Huang) and Pioneer Hero (Mr. Xi Wang).