Multi-Modal and Transfer Learning for UAV Image Segmentation

This project is a research initiative combining two advanced deep learning topics in UAV and remote sensing. It covers (1) cross-domain transfer learning for narrow road extraction from satellite to UAV using embedding alignment, lightweight adapters, and few-shot/zero-shot methods, and (2) multi-modal segmentation integrating satellite, UAV, multispectral, LiDAR/DSM, and radar data with attention-based fusion. GPU resources are required for training, adaptation, fusion, and ablation studies.