SungHo Moon
now Ph.D. Candidate at DGIST
Computer Vision & AI Research Engineer
About
I am a Computer Vision researcher specializing in 3D Reconstruction, Multi-Modal AI, and Object Detection. Currently pursuing my Ph.D. at DGIST, I have extensive experience collaborating with industry leaders including Hyundai Motor Company, ETRI, and the Ministry of National Defense and etc. My research focuses on developing robust AI systems and advancing practical AI technologies that can be directly applied to real-world challenges.
News
- Sep 2025: 🏆 Won 1st Place in ICCV 2025 Amazon Grocery Vision Challenge (TAL & STAL tracks)
- Jul 2025: ✅ Completed projects with Huvitz on real-time 3D reconstruction
- Dec 2024: 💊 Completed projects with ETRI on pill detection and recognition
- Nov 2024: 📷 Completed projects with HD Korea Shipbuilding & Offshore Engineering on camera calibration
Major Projects
ICCV 2025 Amazon Grocery Vision Challenge - 1st Place Winner
Jul 2025 - Aug 2025Amazon (ICCV 2025 Challenge)
Develop a multi-modal AI model for Temporal Action Localization (TAL) and Spatio-Temporal Action Localization (STAL) in grocery shopping scenarios.
Achievement: Achieved 1st place in both TAL and STAL tracks within just 1 month of development. Successfully deployed multi-modal model achieving state-of-the-art performance on Amazon grocery dataset.
Real time 3D Reconstruction using Dental Scanner
Jun 2024 - Jul 2025Huvitz
Develop a real-time 3D reconstruction system using scanner.
Achievement: Improved speed by up to 80% compared to the existing algorithm without performance degradation.
Development of a 3D Pose Estimation and Shape Reconstruction Program for Solid Pharmaceuticals
Sep 2024 - Dec 2024ETRI
Developed a prototype system to estimate 3D pose and reconstruct shapes of solid pharmaceuticals, enabling automatic pill detection, recognition, and counting without additional training.
Achievement: Demonstrated accurate pill classification and counting, showcasing potential for automated pharmaceutical management.
Algorithm Development for Automated Image Processing of Stereo Cameras
Sep 2024 - Nov 2024HD Korea Shipbuilding & Offshore Engineering
To design and implement core algorithms enabling automated image processing for stereo camera systems.
Achievement: Delivered a prototype calibration module and contributed to automation pipeline design. Further technical details remain confidential due to project agreements.
R&D of AI Test and Evaluation Standard Model
Oct 2023 - Jun 2024ROKA Headquarters
Create a standard military training/test dataset and build a baseline AI model for introducing various AI weapon systems in the Army.
Achievement: Established initial standards for the Military Performance Certification Center (including dataset construction, baseline model development, and formulation of various strategies).
View all 10 projects →
Establishment of Test and Evaluation Standards for AI Weapon Systems
Mar 2023 - Jun 2024ROKA Headquarters, U.S. Department of Defense
Develop new testing and evaluation standards for AI weapon systems, which differ significantly from traditional weapon systems.
Achievement: Established initial standards for the Military Performance Certification Center (including dataset construction, baseline model development, and formulation of various strategies).
Military Scientific Surveillance System
Mar 2023 - Sep 2023ROKA Headquarters
Reduce false/missed detections and improve true detections by building an AI-based surveillance system.
Achievement: Reduced false positives by 10% compared to the existing system.
Development of Car Location and Speed Estimation Module Using CCTV Footage
Aug 2022 - Dec 2022ETRI
Develop a module capable of estimating vehicle position and speed solely from CCTV video data.
Achievement: Achieved over 90% accuracy in vehicle speed estimation on the target dataset.
Robust Monocular Camera 3D Object Detection in Various Camera Environments
Mar 2021 - Jun 2022Hyundai
Improve the robustness of monocular camera-based 3D object detection, addressing significant performance degradation caused by varying camera environments.
Achievement: Diagnosed key factors affecting model accuracy and significantly improved performance: Accuracy increased from 20% to 80% for a 3-degree angle variation. Accuracy increased from 1% to 50% for a 5-degree angle variation. Research findings contributed to international patents and publications(CVPRw 2024).
3D Building Exterior Reconstruction
Aug 2020 - Dec 2020KETI
Develop a 3D reconstruction module using monocular images.
Achievement: Successfully built a 3D reconstruction module that processes monocular images to generate 3D structures.
Publications
Rotation Matters: Generalized Monocular 3D Object Detection for Various Camera Systems
SungHo Moon, JinWoo Bae, SungHoon Im
CVPR Workshop 2023, June 2023
Proposed a generalized approach for monocular 3D object detection that addresses performance degradation caused by varying camera orientations and systems.
Industry Collaboration
