Sungha Choi

Senior Staff Research Scientist, Manager
Qualcomm AI Research

Email / CV / LinkedIn / Google Scholar / GitHub

He is a Research Manager at Qualcomm AI Research. Before joining Qualcomm, he worked as an AI Applied Scientist at LG AI Research and spent over a decade at LG Electronics, advancing automotive infotainment technology. He earned his Ph.D. in Computer Science and Engineering (CSE) from Korea University in 2022 under Prof. Jaegul Choo, after a decade in industry following his M.S. and B.S. in CSE at Sogang University under Prof. Jihoon Yang. His research focuses on enhancing AI adaptability, spanning robust urban-scene segmentation, test-time adaptation, and fine-tuning large foundation models for personalized AI and video-language models. His work bridges AI theory and real-world applications, driving advancements in both academia and industry.

News

2/2025: One paper accepted to CVPR 2025 New!
7/2024: One paper accepted to ECCV 2024
7/2023: One paper accepted to ICCV 2023
2/2023: Two papers accepted to CVPR 2023
1/2023: One paper accepted to ICLR 2023
7/2022: One paper accepted to ECCV 2022
9/2021: I joined Qualcomm AI Research.
7/2021: One paper accepted to ICCV 2021 (Oral)
2/2021: One paper accepted to CVPR 2021 (Oral)
2/2020: One paper accepted to CVPR 2020
2/2019: One paper accepted to CVPR 2019 (Oral)

Professional Experience

Qualcomm AI Research, Seoul, S. Korea.
Senior Staff AI Research Scientist (Research Manager), Sep. 2021 - Present
LG AI Research, Seoul, S. Korea.
Applied Scientist in Vision Lab., Dec. 2020 - Aug. 2021
Automotive & B2B Center, CTO Division, LG Electronics, Seoul, S. Korea
Lead Software Engineer in Smart Mobility Lab, Jan. 2007 - Nov. 2020

Publications
(† for corresponding author or project lead)

(C12) CustomKD: Customizing Large Vision Foundation for Edge Model Improvement via Knowledge Distillation
Jungsoo Lee, Debasmit Das, Munawar Hayat, Sungha Choi,† Kyuwoong Hwang, and Fatih Porikli
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2025, Accepted (22.1% acceptance rate).
[PDF]
(C11) Feature Diversification and Adaptation for Federated Domain Generalization
Seunghan Yang, Seokeon Choi, Hyunsin Park, Sungha Choi, Simyung Chang, and Sungrack Yun
European Conference on Computer Vision (ECCV), 2024, Accepted (27.9% acceptance rate).
[PDF]
(C10) Towards Open-Set Test-Time Adaptation Utilizing the Wisdom of Crowds in Entropy Minimization
Jungsoo Lee, Debasmit Das, Jaegul Choo, and Sungha Choi†
International Conference on Computer Vision (ICCV), 2023, Accepted (26.2% acceptance rate).
[PDF]
(C9) EcoTTA: Memory-Efficient Continual Test-time Adaptation via Self-distilled Regularization
Junha Song, Jungsoo Lee, In So Kweon, and Sungha Choi†
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023, Accepted (25.8% acceptance rate).
[PDF]
(C8) Progressive Random Convolutions for Single Domain Generalization
Seokeon Choi, Debasmit Das, Sungha Choi, Seunghan Yang, Hyunsin Park, and Sungrack Yun
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023, Accepted (25.8% acceptance rate).
[PDF]
(C7) TTN: A Domain-Shift Aware Batch Normalization in Test-Time Adaptation
Hyesu Lim, Byeonggeun Kim, Jaegul Choo, and Sungha Choi†
International Conference on Learning Representations (ICLR), 2023, Accepted (31.8% acceptance rate).
[PDF]
(C6) Improving Test-Time Adaptation via Shift-agnostic Weight Regularization and Nearest Source Prototypes
Sungha Choi,† Seunghan Yang, Seokeon Choi, and Sungrack Yun
European Conference on Computer Vision (ECCV), 2022, Accepted (28.4% acceptance rate).
[PDF] [TALK1 & DEMO] [TALK2]
(C5) Standardized Max Logits: A Simple yet Effective Approach for Identifying Unexpected Road Obstacles in Urban-Scene Segmentation
Sanghun Jung,* Jungsoo Lee,* Daehoon Gwak, Sungha Choi, and Jaegul Choo (*: equal contributions)
International Conference on Computer Vision (ICCV), 2021, Accepted as Oral Presentation (3% acceptance rate).
[PDF]
(C4) RobustNet: Improving Domain Generalization in Urban-Scene Segmentation via Instance Selective Whitening
Sungha Choi,*† Sanghun Jung,* Huiwon Yun, Joanne Kim, Seungryong Kim, and Jaegul Choo (*: equal contributions)
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021, Accepted as Oral Presentation (4.7% acceptance rate).
[PDF] [TALK1] [TALK2]
(W1) Towards Lightweight Lane Detection by Optimizing Spatial Embedding
Seokwoo Jung,* Sungha Choi,*† Mohammad Azam Khan, and Jaegul Choo (*: equal contributions)
European Conference on Computer Vision Workshop on Perception for Autonomous Driving (ECCVW), 2020.
[PDF]
(C3) Cars Can’t Fly Up in the Sky: Improving Urban-Scene Segmentation via Height-driven Attention Networks
Sungha Choi, Joanne Kim, and Jaegul Choo
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020, Seattle, WA (22.1% acceptance rate).
[PDF] [TALK]
(C2) Image-to-Image Translation via Group-wise Deep Whitening and Coloring
Wonwoong Cho, Sungha Choi, David Park, Inkyu Shin, and Jaegul Choo
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019, Long Beach, CA, Accepted as Oral Presentation (5.5% acceptance rate).
[PDF]
(J1) A New Ensemble Learning Algorithm using Regional Classifiers
Byungwoo Lee, Sungha Choi, Byunghwa Oh, Jihoon Yang, and Sungyong Park
International Journal on Artificial Intelligence Tools 22(4), 2013.
(C1) Ensembles of Region-Based Classifiers
Sungha Choi, Byungwoo Lee, and Jihoon Yang
IEEE International Conference on Computer and Information Technology (CIT), 2007, Accepted as Best Paper Award (1st prize among 188 accepted papers)
[PDF]

Preprints & Ongoing Research

(P4) Multimodal Embedding and Retrieval (Ongoing, Project Lead)
(P3) Efficient Visual Token Compression for Long Video Understanding (Under Review, Project Lead)
(P2) Understanding Personal Concept in Open-Vocabulary Semantic Segmentation (Under Review, Project Lead)
(P1) Concept-Aware LoRA for Domain-Aligned Segmentation Dataset Generation
Minho Park, Sunghyun Park, Jungsoo Lee, Hyojin Park, Kyuwoong Hwang, Fatih Porikli, Jaegul Choo, and Sungha Choi†
Under Review
[PDF]

Patents

26+ U.S. patent applications, with 20 granted (See CV for details)

Research Interests
(Recently active research topics are indicated in boldface)

AI in the Automotive Industry (C5, C4, C3, W1)
Generative AI, Multimodal Large Language Models, Video-Language Models (P3, P2)
Multimodal Embedding, Composed Image/Video Retrieval
Fine-tuning or Personalization of Large Foundation Models (C12, P2, P1)
On-Device AI: Test-Time Adaptation (C10, C9, C7, C6), Efficient Training (C9), Network Compression (Pruning and Quantization)
Domain Generalization (C11, C8, C4), (Source-Free) Domain Adaptation
Continual learning (C9), Self-Supervised/Unsupervised Learning (C6), Out-of-Distribution Detection/Anomaly Detection (C5)
Urban-Scene (Semantic/Instance) Segmentation/Detection (C5, C4, C3, W1)

Awards and Honors

Excellent Paper Award, Korea University, 2022
Fully-Funded Scholarship from LG Electronics for My Ph.D Study
Creative People Award, LG Electronics, 2014
Best Paper Award, International Conference on Computer and Information Technology, 2007
Fully-Funded Scholarship from LG Electronics for My M.S Study

Repositories

RobustNet ★210+
An official PyTorch implementation of “RobustNet: Improving Domain Generalization in Urban-Scene Segmentation via Instance Selective Whitening” (CVPR 2021 Oral)
HANet ★210+
An official PyTorch implementation of “Cars Can’t Fly up in the Sky: Improving Urban-Scene Segmentation via Height-driven Attention Networks” (CVPR 2020)

Invited Talks

(T7) Tech Seminar at Soongsil University, hosted by Prof. Dahuin Jung, Oct. 2024
Towards Human-Like Adaptation of AI Models (1.5 Hours)
(T6) Tech Seminar at AI Frontiers Summit, hosted by KICS Jul. 2023
Recent Advances on Test-Time Domain Adaptation (30 min)
[TALK]
(T5) Tech Seminar at Korea University, hosted by Computer Vision Lab., Jan. 2022
Toward Robust Urban-Scene Segmentation via Height-driven Attention Networks and Instance Selective Whitening (1 hour)
(T4) Tech Seminar at 42dot, Aug. 2021
My Urban-Scene Segmentation Research (1 hour)
(T3) Tech Seminar at AI Retreat, AI Institute, Seoul National University, Apr. 2021
RobustNet: Improving Domain Generalization (10 min)
[TALK]
(T2) Tech Seminar at Automotive Electronic Technology Workshop (hosted by Hyundai Motors Company), Sep. 2016
Connected Car Services with Smart Device Connectivity (45 min)
(T1) Tech Seminar at the conference hosted by Korean Institute of Electronics and Information Engineers, Jun 2016
Connectivity Technology in the Session of Future Automotive Technology (20 min)