Sunoh Lee

I am a Computer Vision & Robotics researcher at Agency for Defense Development (ADD) in the South Korea. My research focuses on high-level intelligent systems designed to navigate and interact with unseen environments.

Research intersets

✦ Open-World Perception
✦ Open-Vocabulary Interaction
✦ Representation Learning
✦ Path Planning in Unseen Environments

Email / CV / Google Scholar / LinkedIn

History

✦ 2022.04 ~ 2025.05 | Agency for Defense Development (ADD) - Military Service

• Research Officer for National Defense (ROND)

✦ 2018.03 ~ 2022.02 | Gwangju Institute of Science and Technology (GIST) - Bachelor

• Bachelor of Science in Electrical Engineering and Computer Science , Minor in Economics
• GPA: 4.2/4.5 (rank: 1/121)

✦ 2015.03 ~ 2018.02 | Hansung Science High School (HSHS) - High School

• High school for talented students in math and science

Publications

(Equal contributions are denoted by *)

OW-Rep: Open-World Object Detection with Instance Representation Learning
Sunoh Lee*, Minsik Jeon*, Jihong Min, Junwon Seo
preprint
project page / video / arXiv

Open-World Object Detection method enhanced by Vision Foundation Models for better detection performance and effective feature similarity recognition among detected instances.

Projects

	Multi Robot Cooperative Autonomous Driving Agency for Defense Development Developed a method to merge detection results, enabling multiple Unmanned Ground Vehicles (UGVs) to cooperate effectively while in motion, based on ROS2.
	Unmanned Reconnaissance Vehicles Development Agency for Defense Development Developed a method to filter moving objects from LiDAR points using the Iterative Closest Point (ICP) algorithm, enabling robust path planning.
	Autonomous Tunnel Exploitation Agency for Defense Development Developed an unseen object detection method for autonomous robotic exploration in subterranean environments using an IP-camera and ROS2. Extracted generalizable latent features of previously unlearnable objects, enabling the system to recognize similarities between objects and detect hazardous materials.
	Deformable Object Recognition Technology Agency for Defense Development Constructed a dataset to enable the detection of deformable objects, such as animals, which may appear in specific environments like off-road terrain, rather than common objects like humans or vehicles. Developed a voxel-based 3D object detection method using LiDAR to identify deformable objects.
	Image Dataset occluded by various type of wires Gwangju Institute of Science and Technology Built an image dataset of objects occluded by various types of wires and validated it through object classification and segmentation. Conducted performance analysis on object recognition under wire occlusions using various augmentations and data preprocessing techniques.

Design and source code from Jon Barron's website