Frank Yu

I am a second year M.Sc. student in Computer Science at the University of British Columbia (UBC), where I am supervised by Prof. Helge Rhodin. I am expecting to graduate in May 2023 and am currently looking for full-time positions!

Previously, I had the pleasure of interning with a wonderful group of people at Google (Project Starline). I also completed my B.Sc. in Electrical Engineering at the University of Manitoba, where I was also an undergraduate research assistant in Prof. Yang Wang's Lab.

Email  /  CV  /  Google Scholar  /  Github  /  LinkedIn  /  Twitter  /  Photography

  New! June 2022: Started internship (Student Researcher) @ Google (Project Starline)
  New! September 2022: Paper accepted to WACV 2023
  New! September 2022: Demo accepted to SIGGRAPH Asia 2022 (XR)
  New! June 2022: Started internship (Research Intern) @ Google (Project Starline)

  September 2021: A-NeRF accepted to NeurIPS 2021 (Poster)
  May 2021: TA'ing CPSC340: Machine Learning and Data Mining
  March 2021: PCL Paper accepted to CVPR 2021 (Poster)
  January 2021: A-NeRF paper available on arxiv
  November 2020: I will be a student volunteer at NeurIPS 2020
  September 2020: Starting my M.Sc in Computer Science at the University of British Columbia
  August 2020: Paper accepted to ECCV 2020 (Spotlight)
  June 2020: Graduated from the University of Manitoba with a B.Sc in Electrical Engineering

My research interests focus on computer vision and neural rendering. My goal is to develop state-of-the-art technologies that seamlessly bend our world with the virtual world and vice versa.

We present a novel method to reduce latency in neural rendering methods for telepresence by caching and implicitly warping neural network features between timesteps.

We present an analysis-by-synthesis approach for monocular motion capture that learns a volumetric body model and refines the 3D pose estimation of the user in a self-supervised manner.

We propose PCL (perspective crop layer), a set of modular neural network layers that when inserted into MLPs or CNNs will deterministically remove location-dependent perspective effects leading to more precise 3D human pose estimation.

We propose a more realistic problem setting for anomaly detection in surveillance videos and solve it using a meta-learning based algorithm.

We present an open-source telepresence platform capable of rendering on varioius view-dependent displays including, spherical, head-mounted VR, and flat screens.

Led weekly tutorials and office hours. Assisted with assignment and final exam grading.

Credits to Jon Barron for the website design.