Zeen (Harry) Chi

Ray Tracing NURBS Surfaces

ShanghaiTech University, Computer Graphics I (CS171)

A tessellation-free NURBS surfaces ray-tracing renderer.

As a non-analytical surface, high-quality-and-efficiency NURBS surface rendering is a very challenging topic in Computer Graphics. Traditional triangulation (tessellation) method requires a huge amount of calculation and memory to generate and store meshes respectively. To solve this problem, in this project, we implemented a path tracer with global illumination to directly render untrimmed NURBS surfaces without tessellating the surface.

Following the methods proposed by Martin et al., we successfully implemented the renderer with more than 3300 C++ baseline. In details, we applied surface refinement for better rendering quality and efficiency, and constructed a BVH to maintain the surface sub-patches and accelerate the ray-patch intersection process.

We had planned to take on a more difficult problem - trimmed NURBS surface - but due to time constraints, unfortunately only a partial implementation was achieved...

Collaborator: Wenchao Li

Close Project

AI-Agent Chinese Chess

ShanghaiTech University, Artificial Intelligence (CS181)

In this project, we built a Chinese chess game panel for human players, with Pygame for Python and JUCE for C++, respectively. Based on the skeleton, we implemented three algorithms, Minimax Search, Q-Learning, and Monte-Carlo Tree Search, to act as AI chess agents.

With the high efficiency of C++, our AI agents can make relatively good decisions in a short time. We also conducted extensive experiments, including 50 games for each pair of agents, to test and compare the power of these agents.

Collaborators: Zhongxiao Cong, Bingnan Li, Suting Chen and Yifan Qin

Close Project

Long-tailed Recognition in Image Classification

ShanghaiTech University, Introduction to Machine Learning (CS182)

For image classification tasks, even with the help of large-scale deep neural networks, such as CNNs, the model will overfit the head data (more samples) and thus inhibit the ability to generalize to the tail data (few samples) due to the long-tail data distribution problem in large-scale datasets.

We analyzed a very representative paper, Decoupling Representation and Classifier for Long-tailed Recongnition, and re-implemented the two-stage disentangled training method it proposed. Based on the implementation, we tested the effect of such a scheme on ImageNet dataset, and concluded that this method does outperforms the traditional one-stage methods.

Collaborator: Zhongxiao Cong

Close Project

Who is Flying over?

ShanghaiTech University, Introduction to Information Science and Technology (SI100B)

Flight data analysis system.

We implemented a crawler to get flight data from https://zh.flightaware.com/, as well as an LED controller(to display the number of flights), a control panel(on website) and data visualization. The crawler can automatically crawl the data per certain seconds, and LED will be lightened at the same time(based on the quantity of flights). We can modify parameters of crawler and LED via control panel to get data in customized areas, and some parts of the data will be displayed on the website.

There are many great features for our project:

1. Frontend is written in Vue.js, the UI is prettified by Vuetify.
2. We used Redis as a message queue to communicate among modules, avoiding shared mutable states.
3. We used JSON format for data storage, which is cross-language and cross-platform.
4. Frontend and backend applications communicate via RESTful API. Data are stored in .json files. Images are stored in Data URI.
5. Coordinates can be modified via the map on the website, which is more efficient(thanks to mapbox).

Congratulations! Our project was selected as an excellent course project (it is top 1 actually) and presented in the Hall of SIST. This is our poster.

Collaborators: Yihang Liu and Cheng Peng

Close Project

Multi-armed Bandit Analysis

ShanghaiTech University, Probability and Statistics (SI140)

A shallow-to-deep analysis of the multi-armed bandit problem.

This project is about a classical problem in reinforcement learning, named Multi-armed Bandit. The project is generally divided into three parts:

• Simulation. We have to simulate the multi-armed bandit via different algorithms under a set of given parameters and show the output, i.e., the performance of the bandit.
• Analysis. We have to analyze the impact of each hyperparameter of each algorithm. Additionally, we have to analyze an important problem in the multi-armed bandit problem, that is, the exploration-exploitation trade-off problem.
• Design. We have to handle some variants of the classical scenario, including dependency and constraint, and design some algorithms by ourselves to address the novel problem variants.

You can refer to the outline to learn about the background and requirements of this project and refer to the report to look through my project report.

Close Project

Chrome Dino Minigame

ShanghaiTech University, Computer Architecture I (CS110)

Chrome Dino pixel game on Longan Nano development board.

In this project, we implemented a Chrome Dino Runner Game using the Longan Nano development board. We utilized PlarformIO to interact with the development board and support cross-language programming. For example, we implemented most of the framework by C, but some of the auxiliary functions are implemented by RISC-V instruction set. The functions can be invoked bidirectionally.

The game we implemented includes:

• A simple start menu.
• A helping panel demonstrating how to control the T-Rex.
• A mode-selection panel to choose easy mode or hard mode.
• A live score board. The score increases as the T-Rex running.
• Two buttons to control the T-Rex to perform two different actions (jump and lower its head) to cross obstacles.
• Dynamically moved obstacles, including cactus on the ground and pterosaur flying in the sky.
• A invincible mode when the score reaches integer multiples of 100. The T-Rex will not be killed by obstacles under this mode. After the score increases by 50, the mode will temporarily disappear.

Here is the demo video:

Collaborator: Zhongxiao Cong

Close Project