cv

Ph.D. (Advisor: Prof. Jinkyu Kim)

Korea University

Major: AI Applications

Bachelor

Yonsei University

Major: Computer Science

• Operating Systems
• Computer Architecture
• Algorithms
• Data Structures
• Artificial Intelligence
• Computer Graphics
• Object-Oriented Programming
• Theory of Computation

Research Intern – Robot & AD Localization

NAVER LABS

Worked on real-time 3D occupancy prediction for autonomous driving. Developed StreamOcc, an efficient voxel-based method achieving state-of-the-art performance with reduced latency and memory usage.

• Designed StreamOcc, an efficient yet effective 3D occupancy prediction model.
• Achieved real-time performance while reducing memory usage.

Research Intern – Motion Prediction

42dot.

Explored agent interaction modeling and future-guided supervision in trajectory prediction tasks for autonomous driving.

• Developed future-aware distillation for multi-agent trajectory prediction.
• Introduced Lane-guided Attention Module (LAM) for goal-directed reasoning.
• Published work at ICPR 2024.

Intern

Korea Univ. Vision & AI Lab

Worked on 3D object detection and occupancy prediction using multi-camera input.

• Implemented ORA3D, a multi-view 3D detection model.
• Tackled the occupancy prediction challenge with Waymo Open Dataset.

Intern – Face Detection Project

Samsung Electronics

Contributed to attention-based face detection and dataset construction.

• Implemented few-shot face detection algorithm with attention mechanism.
• Designed and augmented a large-scale face dataset pipeline.

Mitigating Trade-off: Stream and Query-guided Aggregation for Efficient and Effective 3D Occupancy Prediction

Seokha Moon, Janghyun Baek, Giseop Kim, Jinkyu Kim, Sunwook Choi

arXiv Preprint

StreamOcc introduces a real-time 3D occupancy prediction framework that combines stream-based voxel aggregation and query-guided refinement to improve both efficiency and accuracy.

BEVMap: Map-Aware BEV Modeling for 3D Perception

Mincheol Chang, Seokha Moon, Reza Mahjourian, Jinkyu Kim

IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

Proposes a map-aware BEV representation that improves 3D perception by incorporating spatial priors from HD maps and camera-map projection alignment.

Learning Temporal Cues by Predicting Objects Move for Multi-camera 3D Object Detection

Seokha Moon, Hongbeen Park, Jungphil Kwon, Jaekoo Lee, Jinkyu Kim

IEEE International Conference on Robotics and Automation (ICRA)

Introduces the Detection After Prediction (DAP) strategy that uses motion prediction as an auxiliary task to help learn temporal cues for 3D object detection from multi-camera input.

Who Should Have Been Focused: Transferring Attention-based Knowledge from Future Observations for Trajectory Prediction

Seokha Moon, Kyuhwan Yeon, Hayoung Kim, Seong-Gyun Jeong, Jinkyu Kim

International Conference on Pattern Recognition (ICPR)

Proposes an attention-guided knowledge distillation framework that uses future observations to help the model attend to the most relevant agents during trajectory prediction.

VisionTrap: Vision-Augmented Trajectory Prediction Guided by Textual Descriptions

Seokha Moon, Hyun Woo, Hongbeen Park, Haeji Jung, Reza Mahjourian, Hyung-gun Chi, Hyerin Lim, Sangpil Kim, Jinkyu Kim

European Conference on Computer Vision (ECCV)

VisionTrap leverages vision-language supervision to enhance trajectory prediction by guiding the model on what to focus on for better understanding of agent behavior.

RUFI: Reducing Uncertainty in behavior prediction with Future Information

Seokha Moon, Sejeong Lee, Hyun Woo, Kyuhwan Yeon, Hayoung Kim, Seong-Gyun Jeong, Jinkyu Kim

CVPR Workshop on Vision-Centric Autonomous Driving (VCAD)

RUFI improves behavior prediction by leveraging future information as privileged knowledge to reduce trajectory uncertainty during training.

ORA3D: Overlap Region Aware Multi-view 3D Object Detection

Wonseok Roh, Gyusam Chang, Seokha Moon, Giljoo Nam, Chanyoung Kim, Younghyun Kim, Jinkyu Kim, Sangpil Kim

British Machine Vision Conference (BMVC)

ORA3D addresses performance drops in overlapping camera views by applying weak depth supervision and feature alignment across overlapping regions.

First Prize in Computer Science

Yonsei University

Awarded first place in the Department of Computer Science for academic excellence.

HUAWEI Scholarship

Talent Development Foundation

Awarded for academic excellence and potential in technology research.