|
Joonwoo Kwon
Ph.D. student in Computer Science and Engineering
Michigan State University (MSU)
Email /
LinkedIn /
CV /
Scholar /
Github
I am a first-year Ph.D. student in Computer Science at Michigan State University,
where I'm fortunate to be advised by Prof. Zijun Cui.
Prior to this, I earned my B.S. in Electronic and Electrical Engineering at SKKU
and my M.S. in Bioengineering at SNU,
where I was advised by Prof. Jiook Cha.
I also had the opportunity to collaborate with Prof. Shinjae Yoo
and Prof. Yuewei Lin at Brookhaven National Laboratory (BNL),
working at the intersection of neuroscience and deep learning.
|
|
Research Statement
My research lies at the intersection of computer vision, generative AI, and physics-informed learning.
The core objective of my work is to embed physical laws into deep generative and vision models to achieve interpretable and
physically coherent 3D motion representations. Specifially, I explore:
- Physical Knowledge Representation — modeling physical principles within AI systems to enable grounded reasoning and interpretable real-world perception.
- Knowledge Integration — incorporating these representations into generative architectures for realistic 3D motion estimation and generation.
Outside of research, I enjoy playing a wide range of sports including football 🏈 and occasionally capturing moments through photography 📸.
|
Exciting News
🚀 (Aug. 2025) Our paper, Revisiting Your Memory, is accepted for the Oral Paper at ACM Multimedia 2025 - CogMAEC Workshop.
🚀 (Apr. 2025) I'll start my Ph.D. in MSU CS, beginning Fall 2025.
🚀 (Oct. 2024) Our team won the Grand Prize at the AI & Art Hackathon and presented the work at the ART DIFFUSION, SNU MoA.
|
|
|
Stylus: Repurposing Stable Diffusion For Training-Free Music Style Transfer on Mel-Spectrograms
H. Wang*, J. Kwon*, S. Kim*, J. Seo, S. Yoo, Y. Lin and J. Cha
arXiv, 2025
[paper]
We present Stylus, a training-free framework that repurposes a pre-trained Stable Diffusion model for music style transfer in the mel-spectrogram domain.
|
|
|
Macro2Micro: Cross-modal Magnetic Resonance Imaging Synthesis Leveraging Multi-scale Brain Structures
S. Kim*, J. Kwon*, J. Kwon*, J. Min, S. Bae,
S. Yoo, Y. Lin and J. Cha
arXiv, 2024
[paper]
We designed an image-to-image translation model based on Generative Adversarial Network for cross-modal MRI synthesis.
|
|
|
Revisiting Your Memory: Reconstruction of Affect-Contextualized Memory via EEG-guided Audiovisual Generation
J. Kwon*, H. Wang*, J. Yi*, S. Kim*,
S. Yoo, Y. Lin and J. Cha
ACM MM CogMAEC 2025 (CogMAEC), Oral Presentation
[paper]
We proposed a novel generation task, dataset, and a multimodal framework for reconstructing video with music contextualized by human affect from brain signals.
|
|
|
AesFA: An Aesthetic Feature-Aware Arbitrary Neural Style Transfer
J. Kwon*, S. Kim*, S. Yoo, Y. Lin and J. Cha
AAAI 2024,
[project page][paper]
, Acceptance Rate: 23.75% (2,342/12,100)
We present AesFA, a lightweight style transfer method that captures aesthetic style through
frequency decomposition and contrastive training—without relying on pre-trained CNNs.
|
Manuscript in Preparation
|
|
|
An Instance-Adaptive Photorealistic Style Optimization for Commercial Image Harmonization
S. Kim*, J. Kwon*, J. Shin, J. Cha and S. Kim
We developed a relighting and harmonization framework that resolves lighting, texture, and color mismatches
in AI-generated commercial imagery.
|
|
|
Compositional Brain Decoding from Symbolic Representations in the Hierarchical Visual System
S. Kim*, J. Kwon*, H. Wang, J. Kwon,
M. Park, S. Yoo, Y. Lin, V. Pavlovic, Z. Cui and J. Cha
We propose a compositional brain decoding framework that maps fMRI signals to spatially grounded noun
phrases by disentangling semantic and spatial pathways, enabling interpretable and controllable
brain-to-image generation without ground-truth supervision.
|
|
|
A Viscosity-guided Artistic Style Optimization via Brushstroke Parameterization
J. Kwon*, S. Kim*, S. Lee, S. Yoo, Y. Lin and J. Cha
We designed viscosity-aware style optimization and brushstroke parameterization to emulate the physical and textural properties of oil painting and watercolor.
|
|
|
My father is a researcher in electrical engineering, and my mother is an artist. In many ways, what I pursue lies at the intersection
of their worlds. The painting on the left is one of her works.
|
|
|
I spent part of my childhood in Pullman, WA, where I completed eighth grade. I was voted "Most Friendly" and
runner-up for "Most Inspirational" in 2011. Go Spartans — miss you all.
|
|
