Ting-Hsun Chi

I am Ting-Hsun (Austin) Chi, an MSEE student at Stanford University. I graduated from National Taiwan University with a B.S. degree in Electrical Engineering. I have been advised by Prof. Homer H. Chen on light field displays, and by Prof. Yu-Chiang Frank Wang on 3D reconstruction. My research interests include diffusion models, generative models, and neural rendering. I have also interned at IBM Research and Inventec AI Center, where I applied my research to real-world problems in AI for EDA and smart manufacturing. I am passionate about advancing the state of the art in computer vision and graphics, and I look forward to contributing to the research community at Stanford and beyond.

Email  /  CV  /  Scholar  /  Github /  LinkedIn

News

- [2025/06] Our paper "Observability of Step-Edge Artifact of Light Field AR Display" received College Student Research Creativity Award issued by National Science and Technology Council.

Research

I'm interested in diffusion models, generative models, and neural rendering.

We unify the strengths of 3D Gaussian Splatting and sparse-voxel rasterization by introducing a smarter voxel initialization strategy that places detail only where needed and accelerates convergence. Combined with an edge-preserving depth-consistency supervision, our method yields more accurate geometry, better fine-structure recovery, and more complete surfaces than prior approaches.

This paper analyzes when light field step-edge artifact become observable, validates the findings with user studies, and offers design guidelines for near-eye light field displays that minimize step-edge visibility while maintaining visual quality.

RTLExplain is an offline method that builds project-specific knowledge bases to significantly improve LLM performance on RTL design tasks like summarization and question answering.

This paper outlines the principles and benefits of our light field display for close-range medical applications and presents an optical method to overcome the spatial-angular tradeoff inherent to light field systems.

We explain the unobservability of split-up effect (now called step-edge artifact) using human visual acuity and attention, provide supporting theory and experiments, and outline design guidelines to reduce the split-up effect in light field displays.

Projects

We propose a multi-concept personalized image editing framework with high-quality images. We introduced "Spotlight Prompting" to guide diffusion models to focus on specific regions and concepts during the editing process.

We propose a monitoring system for door opening and closing. We detect door outer frame with a yolo-v8 model, and extract features with a handcrafted method. Finally, we classify door status with an predefined threshold.

Experience

Developed a semi-automated 3D modeling system that reconstructs laptop models from 2D technical drawings and performs zero-shot port segmentation using vision-language models. Reduced modeling time by over 99%, from roughly two weeks to just three minutes.

Researched in AI for EDA.

Teaching

Website template is adapted from here.