CromSS: Cross-Modal Pretraining With Noisy Labels for Remote Sensing Image Segmentation

We explore the potential of large-scale noisily labeled data to enhance feature learning by pretraining semantic segmentation models within a multimodal fram...