Jeffrey Huang
Research Engineer at Meta
Burlingame, California, United States
Summary
👤
Senior🎓
Top SchoolJeffrey Huang is a research engineer specializing in computer vision and SLAM with nine years of experience delivering optimized algorithms for AR/VR SoCs and commercial products. He spent four years at Qualcomm refining Mapping, Relocalization, and Anchor systems for XR, and now continues research engineering at Meta. His work bridges academic rigor and production constraints—publishing on sparse keypoint-based action recognition using Transformers and implementing efficient two-stream action recognition code that’s publicly used. Jeffrey has a track record of squeezing high accuracy and low compute footprints on mobile hardware (e.g., ultra-low-FLOPs object detectors) and a knack for turning synthetic-data pipelines and keypoint representations into practical, deployable systems.
9 years of coding experience
6 years of employment as a software developer
Master of Engineering - MEng Computer Engineering specialized in Machine Learning, Master of Engineering - MEng Computer Engineering specialized in Machine Learning at Virginia TechBachelor of Science - BS Electrical Engineering, Bachelor of Science - BS Electrical Engineering at National Tsing Hua University
Github Skills (13)
mask-rcnn10
data-handling10
computer-vision10
data-loading10
machine-learning10
pytorch10
faster-rcnn10
action-recognition10
resnet10
dataloader10
python10
load-data10
data-processing10
Programming languages (1)
Github contributions (5)
jeffreyyihuang/two-stream-action-recognition
Jul 2017 - Oct 2019
Using two stream architecture to implement a classic action recognition method on UCF101 dataset
Role in this project:
ML Engineer
ML EngineerContributions:116 commits, 1 PR, 101 pushes in 2 years 3 months
Contributions summary:Jeffrey implemented a spatial CNN architecture for action recognition using the UCF101 dataset, utilizing PyTorch. Their contributions include loading and processing data using dictionary files, defining a custom dataset class, and implementing training and validation loops. The user also integrated a learning rate scheduler to optimize the training process and included functionality to calculate video-level accuracy, which is then recorded.
two-streamucf101methoddatasetrecognition
jeffreyyihuang/pose-based-action-recognition
Jan 2018 - Mar 2018
Contributions:6 commits, 30 pushes, 3 branches in 2 months
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.