Hyunah Ko

I'm an undergraduate student at Yonsei University, majoring in both Civil and Environmental Engineering and Computer Science. I am currently working as an undergraduate researcher at CVLAB, KAIST AI, advised by Prof. Seungryong Kim.

Email  /  CV  /  Scholar  /  Github  /  LinkedIn

Research Interests

My research goal is to develop advanced frameworks for 3D scene understanding and generation that enable (1) physically plausible 3D dynamic synthesis, (2) high-quality neural rendering with accurate material modeling, and (3) seamless integration with generative AI models. Particularly, my interests include:

• Physics-based 3D dynamic scene synthesis
• Diffusion models for 3D content generation
• Neural rendering and 3D Gaussian Splatting

Publications

Highlighted publications and projects demonstrating my specific experience in 3D computer vision. See CV for complete project list.

Proposing a novel method for explicit representation of 3D point cloud attributes using Fourier decomposition.

Improving model adaptation performance through backbone augmented training techniques.

Selected Projects

Team Project, Yonsei Artificial Intelligence
Developed a novel view synthesis system combining CameraCtrl and 3D Gaussian Splatting to generate controllable 3D objects from single images with temporal consistency. Our approach addresses the challenge of preserving fine-grained facial details in human-centric 3D content generation by integrating multi-view diffusion models with 3DGS representation.

Team Project, Yonsei Artificial Intelligence
Built an end-to-end recipe recommendation system that detects ingredients from images using YOLOv5, generates personalized recipes with knowledge-augmented LLM, and creates food images using Stable Diffusion. Implemented a user-friendly interface allowing real-time ingredient detection and customized recipe generation based on dietary preferences and available ingredients.

Team Project, Yonsei Data Science Lab
Created a personalized pet memorial application that generates heartfelt letters and artistic images commemorating beloved pets. Fine-tuned LLaMA3 using LoRA for emotionally resonant letter generation, and integrated Stable Diffusion with ControlNet for style-consistent memorial artwork. Applied style transfer techniques to match various artistic preferences, providing comfort to pet owners through AI-generated personalized tributes.

Team Project, Yonsei Data Science Lab
Analyzed Seoul's late-night mobility patterns and existing night bus data to identify underserved regions during late-night hours. Performed comprehensive exploratory data analysis on night bus ridership data and applied regression modeling to predict late-night transportation demand. Proposed data-driven night bus route designs to improve accessibility for underserved communities, supported by statistical analysis and geospatial visualization. The project demonstrates practical application of data science to address real urban transportation challenges.

Extracurricular activities

Personal Life

Outside of research, I'm passionate about music, especially live band performances. I frequently attend concerts and enjoy discovering new artists. Music helps me stay creative and inspired in both my personal and academic life.

Website template from Jon Barron.