Junkai Huang

Junkai Huang

MSR student at Carnegie Mellon University

Junkai Huang

Hi there:wave:! I’m a MSR student in the Robotics Institute at Carnegie Mellon University, advised by Prof. Fernando De la Torre. I received my Bachelor’s degree in Computer Science and Mathematics with First Class Honors from HKUST, where I was honored to work with Prof. Chi-Keung Tang and Prof. Yu-Wing Tai. I was an exchange student at Vanderbilt University in my final semester as a bachelor.

My research interests lie at the intersection of 3D computer vision, deep learning and robotics. I’m particularly interested in 3D understanding & reconstruction for robotics applications.

I’m looking for a Ph.D position starting from 2025 Fall!

Publications

* indicates equal contribution.

 Instance Neural Radiance Field
Yichen Liu*, Benran Hu*, Junkai Huang*, Yu-Wing Tai, Chi-Keung Tang,
International Conference on Computer Vision (ICCV), 2023
| paper | video | code
NeRF-RPN: A general framework for object detection in NeRFs
Benran Hu*, Junkai Huang*, Yichen Liu*, Yu-Wing Tai, Chi-Keung Tang,
Conference on Computer Vision and Pattern Recognition (CVPR), 2023
| paper | video | code | data

Projects

 Human-centered Large-scale Dynamic 3D Reconstruction
Fall 2023 - Present
We are developing a new 3D gaussian splatting based dynamic 3D reconstruction pipeline for large-scale, 💃 human-centered sports events and spectacles, towards real-time live broadcast applications. This is an ongoing work.
DragGaussian: Point-based 3D Gaussian Manipulation
Spring 2024
Proposed a diffusion model based 3D gaussian editing method that takes 3D point dragging signals as input, uses 2D supervision to manipulate 3D gaussians.
| code
Photon Mapping - Physics-based Rendering Project
May 2024
🏆 Technical award winner of the in-class rendering final competition. I implemented the photon mapping algorithm with NEE light sampling, final gathering and caustic photon mapping.
Neural Style Transfer with Feedforward Network
April 2024
Implemented the 🎨 neural style transfer algorithm which resembles specific content in a certain artistic style. We further took the CycleGAN generator as the feedforward network to train a generalizable neural stylizer.

Awards & Honors

  • HKUST Academic Achievement Award & First Class Honors 2023
  • HKSAR Government Scholarship Fund - Reaching Out Award 2023
  • The Bright Dream Robotics Scholarship ‐ Academic Performance 2023
  • HKUST University’s Scholarship 2019 - 2023
  • HKUST School of Engineering Dean’s List 2019 - 2023

Teaching Assistant

  • TA: MSBD5016 Deep Learning in Computer Vision (PG level) HKUST | Spring & Fall 2022
  • TA: COMP4411 Computer Graphics HKUST | Spring 2022
  • Student Helper: COMP2012 Object-Oriented Programming and Data Structures HKUST | Fall 2021

Leadership