Welcome to Computer Vision & Spatial AI Lab
Advancing the frontiers of vision computing and artificial intelligence
Explore Our ResearchOur Research Areas
3D Visual-Spatial Perception
Learning-based Visual Localization
Our research centers on learning-based visual localization, which aims to infer precise camera poses using deep learning techniques. Among various approaches, we explore Scene Coordinate Regression (SCR)—a method that predicts dense or sparse 3D scene points directly from sensor data like 2D images or LiDAR scans. This enables robust localization by linking 2D observations with the 3D world geometry.
Privacy-Preserving Visual Localization
Developing techniques that enable visual localization while protecting privacy concerns through methods like paired-point lifting and 3D ray clouds.
3D Object Reassembly and Reconstruction
Reconstructing the original shape of heritage ceramics from multiple fragments using advanced 3D modeling and matching techniques.
Multimodal AI for Industrial Anomaly Detection
Generative AI for Anomaly Image Generation
Development of a Few Shot-based AI Algorithm for Augmentation and Generation of Defect Data.
Vision-Based Defect Detection
Developing algorithms for detecting industrial defects and generating synthetic defect images for training robust models.
Video Anomaly Detection
Detecting anomalous human behaviors in video surveillance through unsupervised and semi-supervised learning approaches.
Medical-Spatial AI
Medical Image Segmentation and 3D Reconstruction
CT-based carotid artery segmentation and 3D vessel structure reconstruction for improved medical diagnosis and treatment planning.
Hand Gesture Recognition
Developing hand gesture recognition systems with HoloLens 2-based mixed reality technology for intuitive human-computer interaction.
Physics-Informed AI
Using Physics-Informed Neural Networks (PINNs) for accurate flow velocity prediction in fluid dynamics applications.
Lab News
Recent Publications
GTA-Crime: A Synthetic Dataset and Generation Framework for Fatal Violence Detection with Adversarial Snippet-Level Domain Adaptation
Seongho Kim*, Sejong Ryu*, Hyoukjun You* and Je Hyeong Hong
IEEE International Conference on Image Processing, IEEE ICIP 2025
A Cross-Attention Multi-Scale Performer with Gaussian Bit-Flips for File Fragment Classification
Sisung Liu*, Jeonggyu Park*, Hyeongsik Kim and Je Hyeong Hong
IEEE Transactions on Information Forensics and Security (TIFS, IF 6.3, Categorical JCR < 8.7%)
Join Our Lab
Contact Us
Location:
Engineering Center Annex Unit 415-1
Hanyang University
222 Wangsimni-ro
Seongdong-gu
Seoul, 04763
Republic of Korea
Email:
jhh37 at hanyang dot ac dot kr
Telephone:
02-2220-2489