C3Net: Cross-Modal Feature Recalibrated, Cross-Scale Semantic Aggregated and Compact Network for Semantic Segmentation of Multi-Modal High-Resolution Aerial Images.