CV
Pdf version is attached on the top right
Research Areas
- Computer Vision: Depth Estimation, 3D reconstruction, Vision Language Modles (VLM)
- Digital Images Processing: RGB Images, Magnetic Resonance Images (MRI), Hyperspectrum Images (HSI)
- Edge AI: Light-weight Model Design, Model Optimization, Software-hardware Codesign
Technical Skills
- Programming Languages: Python, Java, C++, MATLAB, JavaScript
- Frameworks/Tools: PyTorch, TensorFlow, OpenCV, SNPE SDK
- Development Platforms: Aria AR Glasses, NVIDIA Orin Jetson Nano, Qualcomm Snapdragon, Qualcomm Automotive Platform (SA-8295)
Education
-
University of California, Irvine - PhD in Computer Engineering
Irvine, USA- Co-advised by Prof. Glenn Healey and Prof. Hyoukjun Kwon
- Thesis: Extending Computer Vision for AR/VR Devices
- Thesis committee: Glenn Healey, Hyoukjun Kwon, Salma Elmalaki
2020 - Present -
RWTH Aachen
Achen, Germany- Germany Smart Industrial 4.0 Winter Camps Study Project
2018 -
Southeast University - Bachelor (Hornor) of Computer Science and Technology
Nanjing, China- Advised by Prof. Youyong Kong
- Graduated with an Honors Degree from Chien-Shiung Wu College, the university’s Honors College.
2016 - 2020
Research Experience
-
Efficient Streaming 3D Reconstruction with XR devices
ILLIXR (Illinois Extended Reality) Lab | UIUC, IL & UC Irvine, CA- Designed a graph-based memory bank mechanism for VGGT, enabling streaming input and achieving up to 6.3x faster inference while maintaining constant peak memory usage
- Modeled frame-to-frame relationships using a graph structure, significantly improving ViT efficiency for AR/VR and long-content scenarios; outperforms the SOTA baseline by 20.7%, 5.9%, and 5.3% in camera pose accuracy, depth accuracy, and 3D reconstruction quality, respectively.
- Submitted one paper under review at CVPR 2026.
Oct. 2024 - Present -
Efficient Depth Estimation with Meta Glasses
Intelligent System Architecture Laboratory | UC Irvine, CA- Proposed hardware-friendly novel efficient model with MultiHead Cost Volume, decreased inference latency by 7% and decreased error rate by 17% comparing with previous model
- Proposed Rectification Positional Encoding (RPE) to enable fast online stereo rectification, reducing inference latency by 43% and lowering the overall pipeline error rate by 42%.
- Implemented models with cross-compiling and quantization skills on Qualcomm Snapdragon devices, Nvidia Orin Jetson, and next-generation AR glasses, Aria
- Published one paper to CVPR 2025
Mar. 2023 - Nov. 2025 -
Air quality prediction with hyperspectrum images by SmolVLM
Irvine Vision Laboratory | UC Irvine, CA- Integrated pressure, humidity, and hyperspectral imaging data to predict ground-level air quality using Vision-Language Models (VLMs), improving multi-modal environmental sensing performance.
Nov. 2025 - Present -
Brain Tissue MRI Segmentation Based on LBP 3D
Laboratory of Image Science and Technology | Nanjing, Jiangsu- Designed novel algorithms for brain tissue semantic segmentation in magnetic resonance images; solved it with Supervoxel, LBP 3D algorithm and KNN algorithm
- Gained one national patent and published one paper to IJML.
Oct. 2018 - June 2020
Work Experience
-
Internship - Digital Cockpit Department
NIO - Electric vehicle manufacturer | San Jose, CA- Acceleration and deployment of large language models on heterogeneous in-vehicle devices
- Optimized language model inference with speculative sampling, achieving a 2.5x speedup.
- Deployed large language models on NVIDIA Orin Jetson Nano and Qualcomm SA-8295P platforms.
- Collaborated with other team members on generative models, including diffusion-based image synthesis.
- Check more details in the project page
Summer 2024
Honors and Awards
- June 2020
- Honor Bachelor Degree of Southeast University
- Sept. 2019
- National Encouragement Scholarship (Top 5%)
- Jan. 2018
- Honorable Mention in Mathematical Contest in Modeling of the U.S. (Top 15%)
- Sept. 2017
- National Encouragement Scholarship (Top 5%)