Who is Hyunseo Koh?

Research Interest

I am a Master’s student in the Department of Metaverse at Soongsil University, working in the Reality Lab.
My research spans Graphics, Vision, and Machine Learning, focusing on modeling and transforming appearance, geometry, and motion across 2D, 3D, and temporal domains.
My work includes 3D Style Transfer, where I explore transferring visual attributes within volumetric or spatial representations; fine-grained latent space–based Image Editing for structural and semantic manipulation; Emotion-Driven Motion Generation to synthesize expressive, human-like movements; and 3D Reconstruction under degraded conditions, particularly removing fog to recover clean scene structures.

I am broadly interested in generative and representation-learning methods capable of robust reconstruction, transformation, and temporal evolution of complex digital environments.

Publications

A Study on the Method of Creating Mixed Images by Improving Sentence-Based Image Diversity and Utilizing Image Characteristics.

Hyunseo Koh*, Min Park, Korea Artificial Intelligence Conference 2024.

Optimization of Atmospheric Scattering–Based Methods for 3D Dehazing

Hyunjun Koh, Hyunseo Koh, Heewon Kim, Korean Institute of Information Scientists and Engineers (IPIU 2025)

FEEL: Focus on Emotion-Enriched Motion Using Unsupervised Learning for Text-to-Motion Generation

Youngjae Choi*, Hyunseo Koh*, Heewon Kim (CVPRW 2026)

Project

Alphabet Sign Language Interpretation System

Lorem ipsum dolor sit amet nisl sed nullam feugiat.

Image restoration technology service based on surrounding information

Lorem ipsum dolor sit amet nisl sed nullam feugiat.

Mustagram: Music recommand social media service

Lorem ipsum dolor sit amet nisl sed nullam feugiat.

Raspberry Pi-based Yolov8 Object Detection, Segmentation and Car Tracking

Lorem ipsum dolor sit amet nisl sed nullam feugiat.

AI_XR-Space SME R&D Technology Development Support Program.

Realitylab & Crosslab collaboration.

Award

Contact

For inquiries, research collaboration, or project discussions, please leave a message below. Provide your name, email, and a brief description, and your default mail client will open with the prepared content.