About Me
Hello! I’m Huili Huang(黄慧丽）. I am a 5th year PhD candidate of Computational Science and Engineering(CSE), I’m advised by Dr. David Frost and Dr. Max Mahdi Roozbahani.
My primary research focuses on post-earthquake infrastructure damage assessment using deep learning, computer vision, and generative AI, with a particular emphasis on ground-level imagery. I develop large-scale datasets (e.g., EID, EIDSeg) and design domain-adaptive models built upon state-of-the-art foundation models to address real-world disaster response challenges.
I am currently actively seeking postdoctoral research opportunities in related areas, including computer vision, Gen AI, disaster resilience, and computational modeling.
My academic trajectory began at the University of Electronic Science and Technology of China (UESTC) where I acquired my B.S. in Software Engineering in 2019. I furthered my knowledge at Georgia Tech, earning an M.S. in Computer Science in 2021.
You can view my resume here:
Publications
Research Experiences
Under the mentorship of Dr. Frost and Dr. Roozbahani, I’m part of the “Disaster Group” dedicated to analyzing natural catastrophes such as earthquakes, hurricanes, and wildfires. We employ a range of tools for data acquisition, encompassing satellites, UAVs, and imagery from social media platforms. My primary research areas include:
- Disaster damage analysis based on multi-source data, especially social media data
- Disaster damage assessment based on social media images and Large Vision Model(LVM)
- Geospatial damage assessment based on self-supervised learning
Under the mentorship of Dr. Roozbahani, centered around the exploration and development of data augmentation algorithms harnessing the power of deep learning techniques. A primary emphasis was placed on self-supervised learning methods. A simple and efficient unsupervised representation learning method named ScaleNet based on multi-scale images is proposed to enhance the performance of ConvNets when limited information is available.
The development of a fire-detection system, extracting critical flame features using dynamic detection. Implemented the system in C++, first achieving 93% accuracy with SVM, then enhancing it to 99% accuracy by integrating YOLOv3.