Hyunsoo Kim

Ph.D. Student, Electrical and Computer Engineering, The University of Texas at Austin

Hyunsoo Kim

I am a Ph.D. student in the Department of Electrical and Computer Engineering at The University of Texas at Austin, advised by Prof. Guandao Yang and Prof. Diana Marculescu. Before joining UT Austin, I received my M.S. in Artificial Intelligence from Korea University, where I was co-advised by Prof. Donghyun Kim and Prof. Suhyun Kim.

My research aims to leverage image and video generation models to solve challenging real-world tasks, especially those related to human perception. I am driven by the philosophy of “Generate to Understand”, echoing Feynman’s dictum, “What I cannot create, I do not understand.”

Beyond content creation, I see generative models as fundamental instruments for perceiving, reasoning about, and understanding the physical world. Just as large language models have grown from producing fluent text into systems that reason and solve problems, I believe vision generative models will become the key to understanding the world we live in.

Email: climba (at) utexas (dot) edu  ·  Curriculum Vitae

news

Aug 2026 Started my Ph.D. at UT Austin, co-advised by Prof. Guandao Yang and Prof. Diana Marculescu.
Aug 2026 Received my M.S. in Artificial Intelligence from Korea University.
Jun 2026 Two papers are accepted to ECCV 2026! See you in Malmö! 🇸🇪
May 2025 Our paper on data-free image synthesis (DDIS) is accepted to ICML 2025! See you in Vancouver! 🇨🇦
Feb 2025 Our paper on image analogy generation (Difference Inversion) is accepted to CVPR 2025! See you in Nashville! 🇺🇸

selected publications

Safety-Aware Image-to-Image Translation without Paired Data preprint

Safety-Aware Image-to-Image Translation without Paired Data

Hyunsoo Kim, Wonjun Lee, Donghyun Kim, Suhyun Kim

Preprint, 2026.

Correlation-Weighted Multi-Reward Optimization for Compositional Generation ECCV 2026

Correlation-Weighted Multi-Reward Optimization for Compositional Generation

Jungmyung Wi, Hyunsoo Kim, Donghyun Kim

European Conference on Computer Vision (ECCV), 2026.

arXiv code project page
Self-Improving Diffusion Classifiers with Minority Preference Optimization ECCV 2026

Self-Improving Diffusion Classifiers with Minority Preference Optimization

Hyunsoo Kim*, Jungmyung Wi*, Soobin Um, Donghyun Kim†, Suhyun Kim†

European Conference on Computer Vision (ECCV), 2026. (*: equal contribution, †: equal advising)

arXiv code project page
When Model Knowledge meets Diffusion Model: Diffusion-assisted Data-free Image Synthesis with Alignment of Domain and Class ICML 2025

When Model Knowledge meets Diffusion Model: Diffusion-assisted Data-free Image Synthesis with Alignment of Domain and Class

Yujin Kim*, Hyunsoo Kim*, Hyunsoo J. Kim, Suhyun Kim

International Conference on Machine Learning (ICML), 2025. (*: equal contribution)

arXiv project page
Difference Inversion: Interpolate and Isolate the Difference with Token Consistency for Image Analogy Generation CVPR 2025

Difference Inversion: Interpolate and Isolate the Difference with Token Consistency for Image Analogy Generation

Hyunsoo Kim, Donghyun Kim†, Suhyun Kim†

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025. (†: equal advising)

arXiv project page

See all publications →

education

experience