Joonwoo Kwon
My research interest lies at the intersection of Computer Vision, Multimodal Understanding and Generative AI.
I aim to integrate multimodal understanding and cross-modal generation techniques to develop AI systems capable of producing context-aware, emotionally resonant, and user-personalized multimedia content.
📌 I plan to apply for Ph.D. programs in Electrical Engineering or Computer Science for the Fall 2025 intake!
Current
Research Scientist at Hanhwa Systems Co., Ltd. (Incoming)
(Previously) Research Associate at Connectome Lab, SNU
Education
- M.S. in Applied Bioengineering at Seoul National University (SNU)
- B.S. in Electronic and Electrical Engineering at SungKyunKwan University
Research Advisor
I've had the privilege of being advised by Prof. Jiook Cha at Seoul National University (SNU), as well as Prof. Shinjae Yoo and Prof. Yuewei Lin at Brookhaven National Laboratory (BNL).
News
Dec. 2024
🔥 Our paper on affect-contextualized audiovisual generation was accepted to the AAAI 2025 Workshop on AI for Music (AI4Music).
Oct. 2024
🔥 We successfully presented our work, "The Recollection of Your Most Cherished Experience Utilizing AI and Neural Signals," at the Tech to Art Platform (TAP) International Conference Prequel – ART DIFFUSION, held at the SNU Museum of Art.
Oct. 2024
🔥 Our team won the GRAND prize 🏅 at the AI & Art Hackathon hosted by the AI Art Research Center in Seoul, Korea! 🎉
© Joonwoo Kwon
Dec. 2023
Research Projects
Previously, I have worked on several research projects spanning from style transfer to brain-to-image decoding.
AesFA: An Aesthetic Feature-Aware Arbitrary Neural Style Transfer
AAAI 2024, First Author; Acceptance Rate: 23.75% (2342/12100)
Developed a new neural style transfer method (AesFA) that effectively utilizes frequency domain decomposition for images to better disentangle aesthetic styles from the reference image.
Macro2Micro: A Rapid and Precise Cross-modal Magnetic Resonance Imaging Synthesis using Multi-scale Structural Brain Similarity
PREPRINT; First Author
Developed a new deep learning framework, Macro2Micro for predicting the microstructure of the brain from the microstructure of the brain while achieving favorable accuracy and speed.
Details are available upon request.
Visual Attention Guidance Enables A Composable Brain-To-Image Decoding
MANUSCRIPT IN PREPARATION; First Author
Inspired by the two-stream hypothesis, we proposed a new brain-to-image decoding framework that directly guides both the placement and identity of the objects to the model, for the first time enabling a composable brain decoding.
Details are available upon request.
AesPHA: An Aesthetic Physics-Aware Neural Style Transfer
MANUSCRIPT IN PREPARATION; First Author
Developed a unique neural style transfer method that effectively captures the style information from the pre-trained fluid-simulation field, utilizing a new style encoding approach that harnesses the chaotic characteristics of the Lorenz system.
Details are available upon request.
Ongoing Projects
Recently, I have worked on projects to develop new generative models and their applications to commercial photography.
An Image Harmonization Model for Visual Coherence in Commercial AI-Generated Photography
Work was done at Planningo (Sept. 2024 - Present)
I aim to develop image harmonization models for the commercial photography service Photio, by resolving the incongruity between AI-generated backgrounds and original photography/videography during synthesis. Specifically, we fine-tuned the synthesis process by matching the lighting conditions of AI-generated backgrounds with those present in the original photography.
An Image Harmonization Model for Visual Coherence in Commercial AI-Generated Photography
Work was done at Planningo (Sept. 2024 - Present)
I aim to develop image harmonization models for the commercial photography service Photio, by resolving the incongruity between AI-generated backgrounds and original photography/videography during synthesis. Specifically, we fine-tuned the synthesis process by matching the lighting conditions of AI-generated backgrounds with those present in the original photography.
The Recollection of Your Most Cherished Experience utilizing AI and Neural Signals
The Grand Prize 🏅
AI & Art Hackathon, AI Art Research Center, SNU (Oct. 2024)
Proposed a multimodal AI framework for synthesizing personalized videography (video+music) using generative AI and neural signals (EEG).
We measured emotional changes in the subject using EEG and injected extracted emotional information into corresponding neural representations. We showed that the model with affect (emotion) created more realistic and high-quality results. Our work will be presented at the Tech to Art Platform (TAP) International Conference Prequel.
Selected Projects
I have had the opportunity to work with several esteemed corporates and teams.
Industrial Projects
Mar. 2022 - Jun. 2022
Samsung Advanced Institute of Technology Research Capstone
Coordinated research on an image-to-image translation model designed to increase semiconductor yield by reducing errors in the etching process. I employed U-NET and PatchGAN to build networks for synthesizing 3D depth maps from Scanning Electron Microscope (SEM) Imaging. The model's effectiveness, measured in Root Mean Squared Error (RMSE), was rated in the top 20% compared to other models.
Dec. 2019 - Feb. 2020
A New Blockchain Management System for the Cement Industry (Smart Factory)
ITECH Industrial Systems, Gwangmeong, Korea
Led a team of three developers to create a new blockchain management system that enhanced the transparency and validity of the cement transportation process, allowing for the real-time monitoring of multiple cement storage silos.
Nov. 2019
SK Telecom Software Engineer Internship (Smart Trash Can)
We introduced a new incentive program that provides users with mileage points that are usable as campus cash as a reward for proper garbage disposal.
Art / AI-Creation Projects
Sept. 2024 - Oct. 2024
AI & Art Hackathon: Towards The Most Personalized Artworks
Created a multimodal AI framework for synthesizing personalized videography using neural signals.
Sept. 2020 - Dec. 2020
An Appreciation Aid Tool For The Visually Impaired Via Synesthetic Perception
Created an appreciation tool for the visually impaired using Python and Arduino; This tool transformed the hue, brightness, and saturation of objects into timbre, pitch, and volume, respectively, of pre-defined musical notes to aid in synesthetic perception.
Sept. 2020 - Dec. 2020
AI & Art: Generation Of New Artworks With Style Transfer
Led a team of five individuals in building a project to transform a modern ballet dance video into the style of Andy Warhol by using Adaptive Instance Normalization (AdaIN) techniques.
Nov. 2019
AI Novel Writing Competition
Worked on a team of three engineers to build an AI model that generates letters/diary entries based on a single word input; We synthesized a 2,000-Korean word personalized diary entry.
Miscellaneous
This is her painting!
© Yoonmee Park
My father is a researcher in electrical engineering, and my mother is an artist. Interestingly, what I pursue is combining the two.
I completed eighth grade at Lincoln Middle School in Pullman, Washington. I was awarded as the most friendly eighth grader and chosen as a runner-up for the 2011 most inspirational eighth grader.
Go Spartans! Miss you all.
If you have any questions, feel free to contact me!
Joonwoo Kwon (pioneers@snu.ac.kr)
Last updated on Dec. 18th, 2024