Junghoon Seo
I lead computer vision and machine learning research for AI robotics at PIT IN Corp.
Experience
Dec 2024 — Present
Executive Director
AI R&D Team Leader · PIT IN Corp.
Sep 2020 — Sep 2024
Technical Leader & Co-founder
Research Center · SI Analytics
Jul 2017 — Feb 2020
ML/CV Research Scientist
Education
Mar 2023 — Feb 2025
M.S. in Culture Technology
KAIST · Daejeon
HCI Tech Lab · Advisor: Prof. Sang Ho Yoon
Mar 2014 — Feb 2021
B.S. in Electrical Engineering and Computer Science
GIST · Gwangju
Specialties
- Machine Learning, Computer Vision, Computer Graphics, and Remote Sensing Applications
- Parallel Computing on HPC and GPGPU
- Intelligent Sensing Techniques for Human-Computer Interaction
Current Focus
- Reliable Perception for AI Robotics
- High-precision Photogrammetry
- AI Agent for User-friendly Automation
- Parallel/Efficient Computing Methods for Robotics
Selected Publications
Variational Approach to Optimal IPS Estimator for Multi-logger Off-Policy Evaluation
NeurIPS
2026
Distributional Stability of Tangent-Linearized Gaussian Inference on Smooth Manifolds
On Pitfalls of RemOve-And-Retrain: Data Processing Inequality Perspective
Thumb Force Estimation with Egocentric Vision
Guided Super Resolution of Land Surface Temperature Using Multi-Satellite Imageries
Hausdorff Distance Matching with Adaptive Query Denoising for Rotated Detection Transformer
Pushing the Limits of Vision-Language Models in Remote Sensing without Human Annotations
A Billion-scale Foundation Model for Remote Sensing Images
Self-Pair: Synthesizing Changes from Single Source for Object Change Detection in Remote Sensing Imagery
Geometric Remove-and-Retrain (GOAR): Coordinate-Invariant eXplainable AI Assessment
Prototype-oriented Unsupervised Change Detection for Disaster Management
Quantile Autoencoder with Abnormality Accumulation for Anomaly Detection of Multi-variate Sensor Data
Contrastive Multiview Coding With Electro-Optics for SAR Semantic Segmentation
Training Domain-invariant Object Detector Faster with Feature Replay and Slow Learner
On the Power of Deep but Naive Partial Label Learning
NL-LinkNet: Toward Lighter but More Accurate Road Extraction with Non-Local Operations
Revisiting Classical Bagging with Modern Transfer Learning for On-the-fly Disaster Damage Detector
Deep Closed-Form Subspace Clustering
Why are Saliency Maps Noisy? Cause of and Solution to Noisy Saliency Maps
Bridging Adversarial Robustness and Gradient Interpretability
RBox-CNN: Rotated Bounding Box based CNN for Ship Detection in Remote Sensing Image
Noise-adding Methods of Saliency Map as Series of Higher Order Partial Derivative
Domain Adaptive Generation of Aircraft on Satellite Imagery via Simulated and Unsupervised Learning
Multi-task Learning for Fine-grained Visual Classification of Aircraft
ACML Workshop
2017
Work in progress
- Pressure Estimation for Hand-based Interaction
- Scalable Multi-robot System
- Data-efficient RGB-D Pose Estimation
- System Identification on Jordan Block
- Lipschitz Bandits under Delay
- Scalable Scheduling for Battery Swapping Service
Awards & Honors
- 2024Lab demo project selected as Popular Choice Winner in CHI 2024
- 2021Best Undergraduate Thesis Award from GIST
- 20205th place in xView2 Challenge
- 20183rd-4th place in CVPR NTIRE Super-resolution Challenge
- 20182nd place in DOTA Challenge
- 2016Grand Prize at KISTI National Supercomputing Competition
- 2016Qualcomm-GIST Innovation Award
Invited Talks
- Query as Representation: A Paradigm Shift in Computer Vision Caused by DETR @ Online. Nov 2022. YouTube (Korean)
- Recent XAI Trends in Deep Learning Era: (Under-)specification and Approaches @ KAERI. Feb 2020.
- Back to the Representation Learning with focusing on Visual Self-supervision @ ETRI. Jul 2019. Slides (Korean)
- Deep Perceptual Super-resolution: Going Beyond Distortion @ KARI. Jul 2018. Slides
Review Services
- Journals: IEEE TPAMI, IEEE TNNLS, IEEE TGRS, IEEE GRSL, and more
- Conferences: AAAI '27, ECCV '26, ICLR '26, CVPR '26, NeurIPS '25, ICML '25, and others
Patents
- Method for detecting object Granted10-2364882
- Method for detecting on-the-fly disaster damage based on image Granted10-2255998
- Method for predicting frame using deep learning10-2023-0115699
- Method, system, and computing device for generating an alternative image of a satellite or aerial image in which an image of an area of interest is replaced10-2023-0075347
- Method of training object prediction models using ambiguous labels10-2022-0000169
- Method for detecting object10-2021-0168545
- Method to detect object10-2021-0135451
- Method for data clustering10-2020-0107206