시각공간인공지능 연구실

This project aims to develop Hyper-CLIP, a multimodal embedding framework that effectively aligns visual and textual semantics on hyperspectral imagery. Building upon this foundation, we further introduce Hyper-LLaVA, optimized for efficient deployment in real-world industrial environments. This project aims to develop a sensor-fusion based Scene Coordinate Regression (SCR) framework for robust outdoor localization, and also explores a federated learning-based collaborative learning pipeline for integrated 3D mapping and localization across multi-device environments. Development of an autonomous foreign object debris detection and digital twin analysis service for aircraft engine ducts using a tethered UGV, integrating diffusion model-based abnormal object generation and detection AI with Vision-LiDAR fused SCR-based localization. This project comprises developing an algorithm for type classification of data sent from geostationary satellites. It involves byte-data augmentation using transformers. This project develops an Intelligent Scene Coordinate Regression (SCR) framework for LiDAR localization that overcomes current issues of storage, privacy, and scalability by introducing dynamic object filtering. The goal of this study is to develop a next-generation humanoid platform capable of bidirectional communication using extrasensory information to autonomously execute human-taught commands and enhance human cognitive and sensory functions. The aim of the project is to create a mixed-reality (MR) platform for calculating smart hemodynamics index in real-time to support patients with cardiovascular diseases. The project is aimed towards developing an accurate algorithm for image-based anomaly detection algorithm that can be applied across a range of different products. W