Computer Graphics Projects
A precomputed dynamic real-time global illumination rendering algorithm
Bidirectional Texture Function Compression
Upsampling Regression Filter
Ultra Large Point Cloud Rendering
A precomputed dynamic real-time global illumination rendering algorithm
- Inspired by Global Illumination with Radiance Regression Functions.
- Objects are dynamic and there can be relative motion among them. We use binary trees of neural networks to simulate the radiance transfer from one object to a certain small surface to compute 1-bounce indirect light.Ÿ
- I wrote the sampling program on GPU and part of the rendering framework, and migrated it to OpenGL-CUDA rendering framework.
Bidirectional Texture Function Compression
- We used neural networks and CBM autoencoder to compress the 6D bidirectional texture function.
Upsampling Regression Filter
- We extracted the core of Joint Bilateral Upsampling with neural networks in order to adapt more general cases.
Ultra Large Point Cloud Rendering
- We are working on a rendering algorithm to implement 3D roaming in scenes of a large set of point cloud. And the data sets are TB-level.
- We use base meshes with level-of-detail height map texture to represent the large scene.
- I proposed an algorithm of computing the level of each triangular face of mesh and finished the design of tree-structured texture attaching and the framework of two-pass rendering.
Tencent Internship Project
Android OCR tool dedicated to recognition of programming codes
Android OCR tool dedicated to recognition of programming codes
- I supervised a 4-member team and design the whole framework.
- I decided to use Tesseract open source library and improved the quality of binarization using Sauvola algorithm.
- I proposed an algorithm using naive bayesian classifier to distinguish lines of code and lines of english text.
- We have applied for three national patents.