Gwanghyun (Bradley) Kim

Welcome!

I am a Ph.D. candidate in electrical and computer engineering (ECE) at Seoul National University (SNU), advised by Prof. Se Young Chun. I completed my M.S. in Bio and Brain Engineering (BBE) at Korea Advanced Institute of Science and Technology (KAIST), advised by Prof. Jong Chul Ye. I earned my B.S degree in electrical and electronic engineering (EEE) from Yonsei University, advised by Prof. Dong Hyun Kim.

Experience

• NVIDIA Research, Research Scientist Intern
  Mentors: Umar Iqbal, Xueting Li, Ye Yuan, Koki Nagano, and Tianye Li
  Jan. 2024 - Feb. 2025, Santa Clara, CA (remote)
• Google Research, Student Researcher
  Hosts: Alonso Martinez, Krishna Somandepalli, and Yu-Chuan Su
  Sep. 2023 - Jan. 2024, New York, NY
• Fast Campus, Lecturer in AI
  Letcure title: Mastering GANs through Model Implementation
  Nov. 2022 - Oct. 2023, Seoul, Korea

Research Interests

My research aims to advance generative AI by building models that genuinely understand the world. I pursue this through three core directions:

• 3D World Understanding: Leveraging 3D geometry to achieve high-fidelity generation, precise control, and robust occlusion handling.
• Multimodal Control: Integrating diverse modalities such as audio, identity, pose, and geometry for nuanced and controllable synthesis.
• Temporal Dynamics: Modeling realistic motion and interactions to ensure temporally coherent content generation.

News

• [Sep. 2025] We won the 2nd place award in REG 2025, the challenge on vision-language foundation models for pathology report generation.
• [Jun. 2025] GeoMan, done at NVIDIA, and PersonaCraft are accepted to ICCV 2025.
• [Oct. 2024] I won the Google Conference Scholarship.
• [Sep. 2024] AVDiT & MoNL, done at Google, is accepted to NeurIPS 2024.
• [Jul. 2024] BeyondScene is accepted to ECCV 2024.
• [Jan. 2024] I joined NVIDIA as a Research Intern.
• [Nov. 2023] I won the Qualcomm Innovation Fellowship South Korea 2023.
• [Sep. 2023] I joined Google as a Student Researcher.
• [Aug. 2023] I am selected as a recipient of Yulchon AI Star Scholarships.
• [Jul. 2023] PODIA-3D is accepted to ICCV 2023.
• [Apr. 2023] Ditto-NeRF is available on arXiv.
• [Feb. 2023] DATID-3D is accepted to CVPR 2023.
• [Nov. 2022] I start a new postiion as Lecturer in AI at Fast Campus.
• [Aug. 2022] DiffusionCLIP is invited to oral session at KCCV 2022.
• [Jun. 2022] DISTL is accepted to Nature Communications.
• [Mar. 2022] DiffusionCLIP is accepted to CVPR 2022.
• [Dec. 2021] I won the Best Master Student for the Year Award at BISPL, KAIST.
• [Dec. 2021] Our AI-based COVID-19 diagnosis solution passed exploratory clinical trial.
• [Nov. 2021] MT-ViT for COVID-19 is accepted to Medical Image Analysis.
• [Sep. 2021] FeSTA is accepted to NeurIPS 2021.
• [Dec. 2020] We won the 2nd place award in NVIDIA AI Healthcare Hackathon.

Research

Patents

• Higher-Resolution Human-Centric Scene Generation With Pretrained Diffusion
  S. Y. Chun, Gwanghyun Kim, H. Kim, H. Seo, D U Kang
  U.S. Patent, Filed, No. 19,019,060, 2025
  Korean Patent, Filed, No. 10-2024-0115484, 2024

• Multi-Modal Diffusion with Mixture of Timesteps
  K. Somandepalli, Gwanghyun Kim, A. Martinez, Y. Su, B. Jou, J. Lezama, A. Gupta, L. Yu, L. Jiang, A. Jansen, J. Walker
  U.S. Patent, Filed, No. 63,548,776, 2024

• Domain Adaptation of 3D Generative Model Across Large Domain Gap Using Pose-Preserved Text-to-Image Diffusion
  S. Y. Chun, Gwanghyun Kim
  U.S. Patent, Filed, No. 18,933,335, 2024
  Korean Patent, Filed, No. 10-2023-0149926, 2023

• Diversity-Preserved Domain Adaptation Using Text-to-Image Diffusion for 3D Generative Model
  S. Y. Chun, Gwanghyun Kim
  U.S. Patent, Filed, No. 18,534,243, 2023
  Korean Patent, Filed, No. 10-2023-0107395, 2023

• Severity quantification and lesion localization method of infectious disease on cxr using vision transformer and apparatus therefor
  J. C. Ye, S. Park, Gwanghyun Kim
  U.S. Patent, Granted, No. 12,169,933, 2024
  Korean Patent, Granted, No. 10-2558096, 2023

• Method of classifying lesion of chest x-ray radiograph based on data normalization and local patch and apparatus thereof
  J. C. Ye, S. Park, Y. Oh, Gwanghyun Kim
  U.S. Patent, Granted, No. 12,056,880, 2024
  Korean Patent, Granted, No. 10-2021-0039509, 2023

Awards and Honors

• 2nd Place Award in Report Generation in Pathology Using Pan-Asia Giga-Pixel WSIs (€400), MICCAI, 2025
• Google Conference Scholarship ($3,000), Google, 2024
• Qualcomm Innovation Fellowship South Korea 2023 ($4,000), Qualcomm, 2023
• Yulchon AI Star Scholarships ($8,000), Youlchon Foundation & SNU AI Institute, 2023
• Brain Korea 21 Scholarships, Korea Research Foundation, 2022 - 2023
• Best Master Student for the Year Award ($2,000), BISPL, KAIST, 2021
• 2nd Place Award in NVIDIA AI Healthcare Hackathon ($8,500), NVIDIA & GIAI, 2020
• KAIST Scholarship, KAIST, 2020 - 2021
• Honors Scholarship, Yonsei University, Spring 2018, Fall 2018

Invited Talks

• BeyondScene: Higher-Resolution Human-Scene Generation With Pretrained Diffusion
  Samsung MX Research Collaboration Seminar, SNU IPAI, 2024
• BeyondScene: Higher-Resolution Human-Scene Generation With Pretrained Diffusion
  Cohere For AI, Cohere (Remote), 2024
• Tutorial on Denoising Diffusion Models
  Bioimaging Winter School, IEIE, 2024
• Audio-Video Latent Diffusion Models
  Google Research, Google, 2024
• Text-driven Control of 2D/3D Image Using Diffusion: DiffusionCLIP & DATID-3D
  Innerverz Seminar, Innerverz (Remote), 2023
• DiffusionCLIP: Text-Guided Diffusion Models for Robust Image Manipulation
  London Machine Learning Meetup, London Machine Learning Group (Remote), 2022
• Diffusion Models for Vision-Language Tasks
  Kakao Brain Open Seminar, Kakao Brain, 2022
• Deep Learning-based Diagnosis of Infectious Diseases on CXR and Audio data
  NVIDIA AI Developer Meetup, NVIDIA (Remote), 2020

Services

• Conference reviewers: CVPR, ICCV, ECCV, NeurIPS, ICLR, ICML, AAAI
• Journal Reviewers: T-PAMI, T-CSVT, ACM Comput Surv

Mentoring

• Heechang Kim (BS@SNU): Text-to-4D Generation, Summer 2023
• Hoigi Seo, Hayeon Kim (BS@SNU): Text-to-3D Generation, Winter 2022
  DITTO-NeRF: Diffusion-based Iterative Text To Omni-directional 3D Model

Extracurricular Activities

• Yonsei Mixed Choir (Avenante), Club president & Section leader, 2014 - 2019

Powered by Jekyll and Minimal Light theme.