Xudong Jiang

I am a final-year MSc student in Computer Science at ETH Zurich.

Previously, I obtained my Bachelor degree in Computer Science and Mathematics at The Hong Kong University of Science and Technology in 2022 and was awarded the Chern Class Achievement Scholarship with outstanding performance.

Email  |  GitHub  |  Google Scholar  |  LinkedIn  |  Twitter

Research

I'm interested in computer vision and machine learning, especially in 3D vision. Currently, I am exploring how neural networks offer new possibilities for foundational problems in 3D vision and how to effectively integrate domain knowledge into these networks.

GLACE integrates pre-trained global and local encodings, enabling scene coordinate regression to scale to large scenes with only a single small-sized network.

Estimate optical flow to align flash and non-flash pair for robust reflection removal in the wild. Monocular depth estimation is used to synthesise realistic misalignment in training.

Other Projects

These include coursework, side projects and unpublished research work.

Optimize two Polybench kernels: symm (compute-bound) and gemver (memory-bound), with three programming paradigms: OpenMP, MPI, and CUDA, achieving comparable performance to SOTA implementations such as AOCL, MKL, OpenBLAS and cuBLAS.

Refine coarse mesh with variational refinement. Optimize integration of photoconsistency cost under silhouette constraints with efficient GPU Monte Carlo integration on sparse voxels around the mesh.

Fast human reconstruction with implicit representation (Ranked top 3 in class). Utilize silhouette carving, depth prior, and importance sampling for fast and accurate reconstruction. Extract mesh with low bias by depth rendering and TSDF fusion

Implemented Position Based Fluid Simulation with Taichi. Simulate multi-phase fluid and fluid-solid interactions with 20k particles in real-time on RTX3080Ti.

Multi-person motion transfer by generating masked videos and alpha compositing them. Utilized coarse masks to reconstruct the background and obtain fine masks with background matting.

Thank Dr. Jon Barron for sharing the source code of his personal page.