Gitaek Kwon 권기택
KO
Gitaek Kwon

Gitaek Kwon 권기택

AI Research Scientist

Computer Vision · Research & Production

I am a computer vision researcher at VUNO Inc., working to bridge the gap between trained models and real-world products.

My primary research interests include learning from limited and imbalanced data, out-of-distribution (OOD) detection, and data augmentation using generative models. Focusing on the medical imaging domain, I have published papers at ICLR, MICCAI, BMVC, and ISBI. I also work across inference optimization and release automation to help research-stage models transition successfully into real-world services. More recently, I have been expanding my research interests by exploring vision-language models (VLMs) and LLM-based agent technologies.

Email GitHub Google Scholar CV

selected publications

All publications →

  • ICLR 2026

    Frequency-Balanced Retinal Representation Learning with Mutual Information Regularization

    S. Lee, S. Kang, I. Park, G. Kwon, J. Baek, D. Park

    Masked autoencoders can overfit low-frequency background structure and underrepresent the high-frequency details needed for recognition. Adds a mutual-information regularizer that aligns encoder representations with high-frequency content without changing the architecture.

    Paper

  • MICCAI 2024

    Semi-supervised Segmentation through Rival Networks Collaboration with Saliency Map in Diabetic Retinopathy

    E. Kim, G. Kwon, J. Kim, H. Park

    Semi-supervised segmentation with sparse pixel labels is vulnerable to noisy pseudo-masks and class imbalance. Trains two networks with different loss weightings, uses the stronger prediction as the pseudo-label, and limits consistency learning to saliency-based reliable regions.

    Paper Code

  • BMVC 2023Co-firstOral

    Improving Out-of-Distribution Detection Performance using Synthetic Outlier Exposure Generated by Visual Foundation Models

    G. Kwon, J. Kim, H. Choi, B. Yoon, S. Choi, K.-H. Jung

    OOD detectors usually need examples from outside the training distribution, which may be unavailable. Uses CLIPseg to locate the in-distribution object and latent diffusion to remove it, turning training images into synthetic OOD samples.

    Paper Code

  • MICCAI 2022First authorChallenge Winner

    Bag of Tricks for Developing Diabetic Retinopathy Analysis Framework to Overcome Data Scarcity

    G. Kwon, E. Kim, S. Kim, S. Bak, M. Kim, J. Kim

    Segmentation and classification had to be learned from limited, imbalanced labels in a modality without public data. Combines deep ensembles and TTA with reliable pseudo-labeling that gradually admits high-confidence predictions.

    Paper Code

  • ICLR 2020

    Generalized Convolutional Forest Networks for Domain Generalization and Visual Recognition

    J. Ryu, G. Kwon, M.-H. Yang, J. Lim

    Random forests struggle to increase each tree's strength without also increasing correlation between trees. Samples triplets from tree split distributions so same-class features move closer and different-class features move apart.

    Paper

awards

Full CV →
  • 2024.05

    1st place · JustRAIGS — AI Glaucoma Screening Challenge ISBI 2024

    322 participants from 37 countries; first overall among 20 teams in the final phase. Two tasks: binary classification of referral need and multi-label classification of ten fine-grained features.

  • 2022.09

    1st place · DRAC22 — Diabetic Retinopathy Analysis Challenge MICCAI 2022

    First place in all three tasks: multi-class segmentation of fine structures, image-quality classification, and severity grading.

  • 2016.11

    Bronze · ACM-ICPC Korea Regional ACM

patents

All patents →

things i built

All posts →