Yu-Ying Yeh

I am a Ph.D. student advised by Prof. Manmohan Chandraker at UC San Diego.

I am working on computer vision problems related to photorealistic content creation for AR applications.

I am fortunate to work with Dr. Kalyan Sunkavalli at Adobe Research and Dr. Ming-Yu Liu at NVIDIA Research during my internship.

Before joining UCSD, I was a research assistant working with Prof. Yu-Chiang Frank Wang at National Taiwan University.

Email  /  CV  /  Google Scholar  /  Github

My research interest mainly focuses on the intersection of computer vision and computer graphics, including but not limited to inverse rendering, 3D reconstruction, material and lighting estimation for indoor scene, object, and portrait. My goal is to enable photorealistic content creation automatically for AR/VR applications. Besides, I am also interested in representation learning, feature disentanglement and video prediction and have done some works related to generative models on videos and domain adaptation.

We propose a single-image portrait relighting method trained with our rendered dataset and synthetic-to-real adaptation to achieve high photorealism without using light stage data. Our method can also handle eyeglasses and support video relighting.

Transfer high-quality procedural materials and lightings from images to reconstructed indoor scene 3D geometry, which enables photorealistic 3D content creation for digital twins.

An open framework which creates a large-scale photorealistic indoor scene dataset OpenRooms from a publicly available video scans dataset ScanNet.

Transparent shape reconstruction from multiple images captured from a mobile phone.

Video generation, interpolation, inpainting and prediction given a set of anchor frames.

A novel and unified deep learning framework which is capable of learning domain-invariant representation from data across multiple domains.

Feature disentanglement for cross-domain data which enables image translation and manipulation from labeled source doamin to unlabeled target domain.

This page is adapted from this template which is created by Jon Barron.