Automatic story generation in digital gaming has been the research top interesting me most. Combining my working experience of developing the cinematic system for Tang Dynasty Warriors  Mobile at Netease Games, and my survey throughout most of the research work addressing this issue, I composed this research proposal reviewing these hitherto most significant work and propose the prototype of a new system driven by the recurrent neural network.  Besides entertainment, this system will hopefully incorporate human intervention to promote the player's mental health or inspire him or her for self-awakening by the customized generated story plot. See the full text of this proposal from here.  

​

This’s a playable demo of an original developed game as my probation period assignment at NetEase Games. It’s developed with by a team of six fresh men of NetEase Games: one game designer, two game artists, two  game programmer including me and a QA. As NetEase Games haven't released this demo, there won't more figures of this project here due to the confidentiality contract I signed with NetEase Games. My contributions to this project include:

​

fighting system

weapons and items system

zombie AI

​

This system is built based on Google Protocol Buffer. Through this system, edits or modifications of models on 3DS Max could get an instant updates and reflections on the game engine side in real -time, including gaming console and game edit tools. Likewise, some edits on the models from game engine side could synchronized to 3DS Max side in real-time. This system helps to shorten the gap between game artist's editing and actual game play, improving relevant work efficiency of game artist and game designer. This prototype system was rebuilt and incorporated into Missah, the next generation game engine of NetEase Games.

This project is my diploma project for the degree of master of science in computer science in HKU. It was graded with the highest level "A" by my department. 

​

Ray tracing is an important theory in the area of computer graphics for producing photo realistic image. In this project, I implemented a ray tracer accompanying with its accelerating algorithm, which is capable to render complex scenes. For instance, three animations from three different scenes offered by BART were produced by my ray tracer. Further more, photon mapping technique was also incorporated in the tracer, enabling it to produce some global illumination effects like caustic. See sample animations rendered by this system from here.
 

This project is my course project for the course of "Geometric Computing" in HKU. The course has been offered mainly for PhD students from the department of computer science in HKU. The project was eventually graded with the highest level "A" among those of other PhD students enrolling in this course. 

Point cloud simplification can help reduce redundant point data in point cloud to produce different levels of detail geometric models. This project studies the main several point cloud simplification methods including their strengths and weakness by comparison. Based on the study, a new point cloud simplification method, that recursively constructing grids structure based on the
largest normal deviation is proposed. Experimental results have shown this method works effectively and efficiently in simplifying point cloud.

This project is my course project for the course of "Advanced Topics in Pattern Recognition" in HKU. The course has been offered mainly for PhD students from the department of computer science in HKU. The project was eventually graded with the highest level "A" among those of other PhD students enrolling in this course. 

Convolutional Neural Networks (CNNs) are variants of Neural Networks, which exploit spatially local correlation by enforcing a local connectivity pattern between neurons of adjacent layers. In this project, I implemented the convolutional neural network for handwritten digits recognition. The MNIST database is used both in training and testing. For accelerating the training speed, C++ AMP technique was used to invoke GPU's computation capacity.  

This project is my diploma project for the degree of bachelor of engineering in computer science & technology in Beijing Jiaotong University.  It was eventually graded with the highest level "A"(top 15%) among those of my fellow students in the department of computer science.  Besides, some of my previous research work involved in this project was published in the "Beijing University Student Innovation Program" in 2010.

​

This project studied two representative feature points and a feature region in image: Harris points, SIFT points and Canny outline. The algorithms of extracting the two kinds of feature points were implemented and further adopted into image matching. Canny feature, on the other hand, was  studied with its nature and its potential use in pattern recognition and classification. Based on the study, I designed and implemented an image retrieval algorithm for those objects characterized by their shape rather than their ambiguous  color feature. For instance, my method was used in gun image retrieval and the experimental results have verified its effectiveness and accuracy.