I am a final-year CS Ph.D. student in the Computer Graphics and Vision Lab at Portland State University. I'm advised by Dr. Feng Liu. Before that, I recieved my Bachelor's degree from Nanjing University. My research interests include data-driven novel view synthesis for immersive rendering, such as ray tracing and volumetric rendering. I am also interested in efficient video transformers. I was mentored by Dr. Zhang Chen and Dr. Zhong Li.

email  /  google scholar  /  github  /  cv  /  linkedin

We extend Gaussian splatting with temporal opacity and polynomial parametric motion/rotation, along with feature splatting to achieve a compact model size. Additionally, new Gaussians are sampled in regions of error to enhance accuracy. Our model supports 8K rendering at 60 FPS, with fish-eye projection, efficient initialization of sparse points, and aggressive pruning of unnecessary points to optimize performance.

We decouple temporal and channel features using transposed attention to focus on channels, and implement a global query strategy to capture global information. Additionally, a depth shift module is introduced to better integrate cross-channel and temporal information, enabling efficient and high-quality video prediction.

We apply next frame estimation for fast monte carlo rendering and estimate error mask to sample the rest pixels with the renderer. We introduce two large-scale animated Ray Tracing datasets with over 60K mega-pixel frames with ground truth optical flows and other G-Buffers produced from UE4 and Blender Cycles.

We accelerate Monte-Carlo rendering via super-resoultion guided by high-resolution fast-to-compute auxiliary features. We build large scale Ray Tracing Image datasets from more than 1000 scenes with ground truth buffers at different scales.