PsSAR-v1.0: Unlocking SAR Panoptic Segmentation

Panoptic segmentation technology can simultaneously acquire information of objects and background, and thus holds extremely high application value in remote sensing image interpretation. However, the development of this technology in the field of synthetic aperture radar information processing is limited, with the core issue lying in the lack of high-quality public datasets. To address this gap, this study constructs the first SAR panoptic segmentation dataset, PsSAR-v1.0. Based on the high-resolution SAR ship dataset, this dataset explores and proposes a standard workflow for label production of panoptic segmentation datasets, and realizes the joint annotation of ship targets and sealand backgrounds for the first time. With excellent compatibility, the dataset can be adapted to open source panoptic segmentation toolboxes (e.g., mmdetection and detectron2). The dataset contains 5,604 sets of images and their corresponding panoptic segmentation annotations, with the total number of annotated pixels reaching 3.6×10⁹. In addition, this paper establishes a performance benchmark based on this dataset, providing a reliable baseline reference for the academic community. This research effectively breaks through the bottleneck of data scarcity and is expected to further promote the development of SAR panoptic segmentation technology.