Computer Vision

This project aims to develop a 6G Digital Twin for handover prediction by constructing a detailed digital replica of a scene using LIDAR scans, and employing sophisticated scene understanding methods to enhance the accuracy and efficiency of digital twins in 6G networks.

Developed an advanced system for holistic 3D scene understanding from a single RGB image, using attention mechanisms to accurately predict object shapes, poses, and room layout.

Validated the effectiveness of self-supervised learning in achieving state-of-the-art results with limited label availability, revolutionizing the approach to 3D object part segmentation.

Employed self-supervised segmentation techniques to autonomously learn intricate scene understanding tasks with limited annotations, specifically in medical contexts such as surgical operating rooms.

Utilized Lidar as an additional depth constraint for Neural Radiance Fields (NeRF) on outdoor scenes, expanding the field of view and improving scene understanding in autonomous driving applications.

This thesis addresses the challenge of weakly supervised point cloud semantic segmentation by leveraging multi-modal information and introducing novel pseudo-labeling techniques.