Recurrent Sinusoidal INRs for Efficient High-Fidelity Representation: Harmonic-line Spectrum Perspective
Expanding a neural representation’s frequency range by reusing one sinusoidal block—without adding parameters.
1st-year Ph.D. student Korea University
Hyunmin Cho
I explore visual intelligence, with a current focus on representation and generation.
I’m in the Integrated M.S.–Ph.D. program at Korea University, working in the Image Processing Algorithm Lab, advised by Kyong Hwan Jin. My current work spans generative models, controllable generation, and neural representations.
Publications
Research
Balancing Fidelity and Diversity in Diffusion Models via Symmetric Attention Decomposition: Hopfield Perspective
Decomposing attention into symmetric energy and skew-symmetric circulation to control image fidelity and diversity.
TAG: Tangential Amplifying Guidance for Hallucination-Resistant Sampling
Amplifying tangential sampler updates to reduce hallucinations, without retraining or extra denoiser evaluations.
Reference-based Super-Resolution via Image-based Retrieval-Augmented Generation Diffusion
Retrieving visual references to guide diffusion toward realistic, detail-preserving super-resolution.
In collaboration with Samsung Research.
ReAL: Reference-to-Image (R2I)-Aware Latent Diffusion for Image Super-Resolution
Recovering image structure and texture with reference features cached and reused throughout denoising.
PaperProject & code forthcoming
Towards Lossless Implicit Neural Representation via Bit Plane Decomposition
Reconstructing digital signals exactly by learning their individual bit planes.
* Equal contribution † Corresponding author
Work Experience
OMRON SINIC X Corporation
オムロン サイニックエックス株式会社
Research Intern · Full-time · On-site
Hongo, Bunkyo-ku, Tokyo
東京都文京区本郷
Upcoming
Approx. –
A few updates
Recent
Two papers at ECCV: recurrent sinusoidal representations and reference-aware super-resolution.
Two papers at ICML: Tangential Amplifying Guidance and attention decomposition for diffusion.
Gold Prize at the 38th Workshop of Image Processing and Image Understanding (IPIU). Certificate
Background & experience
Education
Mar 2024–present
Korea University
Integrated M.S. & Ph.D. in Electrical Engineering
Generative models, world models, and Hopfield networks.
Advisor: Kyong Hwan Jin
Mar 2020–Feb 2024
Gachon University
B.S. in Software
Image super-resolution.
Advisor: Kiho Choi (now at Kyung Hee University)
Research collaborations
Jan–Dec 2026
Samsung Research
Next Generation Display Lab
Uncertainty-Aware Temporal Relation Reasoning under Missing Video Evidence
Jun 2024–May 2025
Samsung Research
Next Generation Display Lab
Development of user-preferred image-based image quality technology
Patent application
Method and Apparatus for Lossless Implicit Neural Representation Based on Bipolar Vector Labeling and Recursive Single Weight Operation
Hyunmin Cho, Kyong Hwan Jin, Yongjun Lee, Jiwon Kim, Woo Kyoung Han
Korea Patent Application No. 10-2025-0150878