Although significant advances have been made in the semantic segmentation of high-resolution remote sensing (RS) images, obtaining accurate pixelwise annotations remains resource-intensive. We propose ...