Spatial Intelligence Lab. in CAU Logo


The Spatial Intelligence Lab at Chung-Ang University (CAU) advances research and education in computer science and artificial intelligence, focusing on 3D vision, machine learning, robotics, and multi-modal large language models (LLMs), aiming to augment human capabilities/intelligence.

Latest News

Sihyeong presented at LG AI Research Tech Talk
μ€‘μ•™λŒ€ν•™κ΅ 곡간 지λŠ₯ 연ꡬ싀 학뢀생 연ꡬ 인턴 λͺ¨μ§‘!! (Undergraduate Research Internship @ Spatial Intelligence Lab, CAU)
Dr. Lee joins Chung-Ang University (CAU) at Fall semester 2025
Jongmin presented a paper at NeurIPS 2024 in Vancouver

Research Interest

Computer Vision & Visual Geometry
Developing robust methods to perceive, understand, and reconstruct the 3D world from visual data.
Computer Vision & Visual Geometry
Multi-Modal LLMs & 3D Visual Foundation Model
Integrating vision and language to build powerful multimodal foundation models for diverse real-world tasks.
Multi-Modal LLMs & 3D Visual Foundation Model
Geometric Representation Learning
Building deep learning models that respect geometric structure through invariance and equivariance principles.
Geometric Representation Learning
Robotics for Real-World Physical AI
Combining spatial perception and machine learning to enable reliable, generalizable embodied AI in the physical world.
Robotics for Real-World Physical AI
Application Research β€” AR/VR & Autonomous Driving, and more.
Translating computer vision and multimodal AI research into real-world systems, from AR/VR to autonomous driving.
Application Research β€” AR/VR & Autonomous Driving, and more.
See all

Latest Preprints & Research

Exaone 4.0 VL: Vision-Language Foundation Model for Enterprise AI Agent
Exaone-4.0-VL, a 32B multi-modal foundation model, achieves SOTA across vision-language benchmarks and empowers enterprise AI agent applications.
Video Announcement Video
Exaone 4.0 VL: Vision-Language Foundation Model for Enterprise AI Agent
3D Equivariant Pose Regression via Direct Wigner-D Harmonics Prediction
SO(3)-equivariant pose harmonics enable consistent single-image 3D orientation estimation by directly regressing Wigner-D coefficients in the frequency domain, overcoming discontinuities in spatial parameterizations.
PDF Cite Code Poster Video
3D Equivariant Pose Regression via Direct Wigner-D Harmonics Prediction
Learning Rotation-Equivariant Features for Visual Correspondence
This work proposes a self-supervised framework using group-equivariant CNNs and a novel group-aligning technique to learn discriminative, rotation-invariant local descriptors for robust image matching and pose estimation.
PDF Cite Code
Learning Rotation-Equivariant Features for Visual Correspondence
Self-Supervised Equivariant Learning for Oriented Keypoint Detection
This work introduces a self-supervised framework with rotation-equivariant CNNs and a dense orientation alignment loss to detect robust oriented keypoints, achieving state-of-the-art results in image matching and camera pose estimation.
PDF Cite Code
Self-Supervised Equivariant Learning for Oriented Keypoint Detection
Self-supervised Learning of Image Scale and Orientation Estimation
This work presents a self-supervised histogram alignment framework for estimating scale and orientation of image patches, achieving superior pose estimation and boosting image matching and 6-DoF camera pose estimation.
PDF Cite Code
Self-supervised Learning of Image Scale and Orientation Estimation
See all