Video Understanding for Activities of Daily Living
Temporal action detection, dense activity recognition, and long-video reasoning over unedited, real-world sequences.
Charlotte Vision Lab
The Charlotte Vision Lab is a computer vision research group at UNC Charlotte focused on building systems that understand and act in the visual world. Our work spans video understanding, multimodal learning, robotic perception, generative modeling, and trustworthy machine vision. Our goal is to build trustworthy systems that can perceive, reason, and assist in complex real-world environments.
Our Research Directions
Vision-Language and Vision-Language-Action Models
Visual question answering, domain adaptation, interpretable decision-making, and embodied reasoning.
Reliable 3D and Generative Vision
3D scene understanding, controllable image generation, and uncertainty estimation in open-world settings.
Highlights
ECCV 2026
VisCoP
Visual probing for efficient domain adaptation of vision-language models with minimal changes to pretrained parameters.
View paperECCV 2026
Ego2Exo VLM
Learning egocentric cues from exocentric video to improve vision-language understanding of daily living activities.
View paperECCV 2026
3D FaceShell
A privacy-preserving framework that controls how vision-language models interpret 3D face avatars while preserving identity.
View paperCVPR 2026
MS-Temba
A Multi-scale Mamba-based Temporal Modeling Framework for efficient temporal action detection in long untrimmed videos.
View paper