Abstract: We explore the potential of large-scale noisily labeled data to enhance feature learning by pretraining semantic segmentation models within a multimodal framework for geospatial applications ...