Hah Min Lew

Ph.D. student at Korea University

Ph.D. student at the Visual & General Intelligence (VGI) Lab, Korea University, supervised by Prof. Gyeong-Moon Park. Current research focus lies in Vision-Language-Action (VLA) models, Vision-Language-Navigation (VLN) models, and multimodal generative models. Recipient of the Presidential Science Scholarship funded by the Ministry of Science and ICT.

Prior to academia, professional experience includes over three years as an AI Researcher at Klleon AI Research. Key contributions involved photo-realistic 3D human generation, enabling lifelike avatar movements and lip-sync. Holds both M.S. and B.E. degrees from DGIST, with a strong foundation in Deep Learning and Signal Processing, advised by Prof. Jae Youn Hwang.

Conference Publications

  • [C5] AudioAvatar: Personalized Audio-driven Whole-body Talking Avatars [paper] [poster]
    Seungeun Lee, SeungJun Moon, Hah Min Lew, Ji-Su Kang, and Gyeong-Moon Park
    CVPR 2026

  • [C4] SMAGA: Secondary Motion-Aware 3D Clothed Gaussian Avatars from Monocular Videos [paper] [poster]
    Seungeun Lee, SeungJun Moon, Hah Min Lew, Ji-Su Kang, and Gyeong-Moon Park
    ICLR 2026

  • [C3] GeoAvatar: Adaptive Geometrical Gaussian Splatting for 3D Head Avatar [paper] [project page] [code]
    SeungJun Moon*, Hah Min Lew*, Seungeun Lee, Ji-Su Kang, and Gyeong-Moon Park (*Equal contribution)
    ICCV 2025

  • [C2] Towards High-fidelity Head Blending with Chroma Keying for Industrial Applications [paper] [project page]
    Hah Min Lew*, Sahng-Min Yoo*, Hyunwoo Kang*, and Gyeong-Moon Park (*Equal contribution)
    WACV 2025

  • [C1] CSS-Net: Classification and Substitution for Segmentation of Rotator Cuff Tear [paper]
    Kyungsu Lee, Hah Min Lew, Moon Hwan Lee, Minji Kang, Junyoung Kim, and Jae Youn Hwang
    ACCV 2022

Journal Publications

  • [J6] Multi-task and Few-shot Learning-based Fully Automatic Deep Learning Platform for Mobile Diagnosis of Skin Diseases [paper]
    Kyungsu Lee, Thiago Coutinho Cavalcanti, Sewoong Kim, Hah Min Lew, Dae Hun Suh, Dong Hun Lee, and Jae Youn Hwang
    JBHI (IF: 7.021) 01.04.2023.

  • [J5] Deep learning-based framework for fast and accurate acoustic hologram generation [paper]
    Moon Hwan Lee, Hah Min Lew, Sangyeon Youn, Tae Kim, and Jae Youn Hwang
    TUFFC (IF: 3.267, Featured on the front cover) 11.24.2022.

  • [J4] Speckle Reduction via Deep Content-Aware Image Prior for Precise Breast Tumor Segmentation in an Ultrasound Image [paper]
    Haeyun Lee, Moon Hwan Lee, Sangyeon Youn, Kyungsu Lee, Hah Min Lew, and Jae Youn Hwang
    TUFFC (IF: 3.267) 08.26.2022.

  • [J3] Intelligent Smartphone-based Multimode Imaging Otoscope for the Mobile Diagnosis of Otitis Media [paper]
    Thiago Coutinho Cavalcanti, Hah Min Lew, Kyungsu Lee, Sang-yeon Lee, Moo Kyun Park, and Jae Youn Hwang
    Biomedical Optics Express (IF: 3.562, Spotlight on Optics) 12.01.2021.

  • [J2] Ultrasonic Blood Flowmeter with a Novel Xero Algorithm for a Mechanical Circulatory Support System [paper]
    Hah Min Lew, Heean Shin, Moon Hwan Lee, Sangyeon Youn, Hee Chan Kim, and Jae Youn Hwang
    Ultrasonics (IF: 4.062) 08.01.2021.

  • [J1] Forward-looking Multimodal Endoscopic System based on Optical Multispectral and High-frequency Ultrasound Imaging Techniques for Tumor Detection [paper]
    Jihun Kim, Hah Min Lew, Jung-Hee Kim, Sangyeon Youn, Hasan Al Faruque, An Na Seo, Soo Yeun Park, Jin Ho Chang, Enjoo Kim, and Jae Youn Hwang
    TMI (IF: 11.037) 02.01.2021.

International Conference Presentations & Proceedings

  • Deep Learning-based Quantitative Diagnosis of Early Dental Caries Using Multimode Image Datasets
    Hah Min Lew, Thiago Coutinho Cavalcanti, Hong Keun Hyun, and Jae Youn Hwang
    SPIE 2022

  • Monitoring of Spatially Selective Calcium Response to Holographic Ultrasound Stimulation Using Fluorescence Imaging
    Moon Hwan Lee, Hah Min Lew, Jae Seong Kim, Sangyeon Youn, and Jae Youn Hwang
    SPIE 2022

  • Smartphone-based Multimode Imaging for Quantitative Diagnosis of Early Dental Caries
    Hah Min Lew, Sewoong Kim, Thiago Coutinho Cavalcanti, Hong Keun Hyun, and Jae Youn Hwang
    SPIE 2021

  • Multimode 3D and Spectral Imaging and Analysis of the Tympanic Membrane Using a Smartphone-based Otoscopic System
    Thiago Coutinho Cavalcanti, Hah Min Lew, Kyungsu Lee, and Jae Youn Hwang
    SPIE 2021 (Oral)

  • Intelligent Wearable Bladder Monitoring Device with an Inertial Measurement Unit for Robust Measurement of Bladder Volume
    Moon Hwan Lee, Hah Min Lew, and Jae Youn Hwang
    IUS 2020

  • Enfaced Multimodal Endoscopic System based on Multispectral and High-frequency Ultrasound Imaging for In situ Tumor Characterizations
    Jihun Kim, Hah Min Lew, Jin-Hyung Park, Jin Ho Chang, and Jae Youn Hwang
    SPIE 2020 (Oral)

Preprint

  • FacePlex: Full-Duplex Joint Speech-Facial Motion Generation for Conversational Avatars [paper] [project page]
    Habin Lim*, Jae-Ho Lee*, Hah Min Lew*, Ji-Su Kang, and Gyeong-Moon Park (*Equal contribution)
    arXiv 2026
  • Deep Learning-based Synthetic High-Resolution In-Depth Imaging Using an Attachable Dual-element Endoscopic Ultrasound Probe [paper]
    Hah Min Lew*, Jae Seong Kim*, Moon Hwan Lee, Jaegeun Park, Sangyeon Youn, Hee Man Kim, Jihun Kim, and Jae Youn Hwang (*Equal contribution)
    arXiv 2023

Education

  • Ph.D. in Artificial Intelligence, Korea University, Mar. 2026 - present

  • M.S. in Electrical Engineering & Computer Science (EECS), Daegu Gyeongbuk Institute of Science and Technology (DGIST), Mar. 2019 - Aug. 2022

  • B.E. in Bachelor of Engineering, Daegu Gyeongbuk Institute of Science and Technology (DGIST), Mar. 2014 - Feb. 2019 (Best Project Award)

Academic Services

  • Journal Reviewer
    • Medical Physics

Work Experiences

  • AI Researcher, Klleon AI Research, Seoul, South Korea, Aug. 2022 - Feb. 2026
    • Accelerated inference using a training-free diffusion sampler (↑4.11×).
    • Developed audio-driven 3DMM generation for virtual avatars with natural human-like expressions and movements. (LVE ↓27.5%, FDD ↓28.9%, MEE ↓27.1%, CE ↓24.1%, Diversity ↑17.7%)
    • Developed a photorealistic head rendering model using Gaussian Splatting, outperforming 5 state-of-the-art models. (MSE ↓59.96%, PSNR ↑4.41dB, SSIM ↑3.85%, LPIPS ↓38.16%)
    • Developed a Head Swap AI model with 212.7x speedup, reducing annual GPU costs by 99.53% (from $2.2M to $10.5K).
    • Built large-scale multimodal data pipelines (9.41M+ frames from in-the-wild videos).
    • Integrated external APIs (NVIDIA Omniverse Audio2Face, OpenAI ChatGPT, TTS) into a streaming avatar system, achieving 25-28 FPS performance.
  • Graduate Researcher, Multimodal Biomedical Imaging and System Lab, Daegu Gyeongbuk Institute of Science and Technology (DGIST), Daegu, South Korea, Mar. 2019 - Aug. 2022
    • Advisor: Prof. Jae Youn Hwang
    • 6 SCIE publications, 7 international conferences, 9 projects, 4 patents, and 2 awards.
    • Designed machine learning-based anomaly detection systems and generative models for biomedical imaging, focusing on multimodal, class-imbalance, and multi-task learning.
    • Developed hardware-software integrated systems for application-specific use cases.
    • Collaborative research with medical doctors from hospitals, including SNUH, SNUDH, and Yonsei Severance.
  • Co-founder, LANTERN, Daegu, South Korea, Nov. 2016 - July 2017
    • Founded a data-driven personalized tutor matching service company in collaboration with Class101.
    • Designed a matching database system and established tutor evaluation metrics for personalized recommendations.