Zijing (Jing) Huang

apart from them, there is me,

cradled in the steadfast grace

a humble fate has given.

黄紫晶

除了他们还有我

享受着朴素命运带来的

一心一意

About

Hi, I’m Zijing (Jing) Huang — a Ph.D. researcher at the University of Florida working with Dr. Won Suk Lee. My research focuses on computer vision, and AI-driven perception systems that help machines see and make better decisions in the real world.

Before starting my Ph.D., I earned my M.S. in Mechanical and Aerospace Engineering at UF and my B.E. in Mechanical Engineering from Beijing University of Chemical Technology, where I founded the university’s baseball team and club. Along the way, I’ve published multiple first-author papers, built practical AI systems for precision agriculture, and earned recognition for research that works both in the lab and in the field.

Outside of research, I enjoy photography, music, and workout. Fred again.. is my favorite artist.

Good taste is all you need :)

Research

Efficient 2D Vision for Detection, Segmentation, and Multimodal Forecasting

Developed a lightweight object detection model integrating a Mamba state space block and frequency-domain channel mixer for in-field crop ripeness staging. Designed an annotation-free segmentation pipeline combining YOLO and SAM, outperforming supervised baselines. Built a multimodal time-series forecasting model that fuses vision, weather, and spatial data for accurate yield prediction in precision agriculture.

Publication

3D Reconstruction and Segmentation for Plant Canopy Analysis

Created an end-to-end pipeline using Planar-based Gaussian Splatting Reconstruction (PGSR) with automated SAM-based 3D segmentation, producing high-fidelity point clouds for canopy volume estimation. Designed a CLIP + Depth-Anything v2 feature fusion approach for robust multi-view reconstruction in COLMAP. Currently developing a transformer-based pipeline for lightweight superquadric 3D object representation and segmentation.

Publication

Robotics Perception and 6-DoF Pose Estimation

Engineering a real-time detection and localization pipeline for a quadruped robotic platform using Visual Geometry Grounded Transformer (VGGT) for 6-DoF fruit pose estimation in 3D space, enabling precise robotic manipulation in unstructured environments.

Multimodal Automatic Target Recognition

Designed a Siamese network with frozen DINOv2 backbones for IR and RGB imagery, enhanced via contrastive learning. Applied Bayesian network decision-level fusion to integrate classifier outputs, improving robustness in multimodal target recognition tasks.

Albums

Contact

Drop me a line at zijinghuang97@gmail.com — I'd love to talk about new projects or collaborations.