RSAR
收藏资源简介:
RSAR是由南开大学、上海人工智能实验室和深圳福田NKIARI联合构建的旋转SAR目标检测数据集,是目前该领域最大的多类别数据集。该数据集包含95,842张SAR图像和183,534个标注实例,涵盖六种典型的SAR目标类别。数据集的构建过程通过弱监督模型生成伪旋转框标签,并经过人工校准,确保了标注的准确性。RSAR数据集旨在解决旋转SAR目标检测领域缺乏大规模数据集的问题,提升模型在角度预测上的精度,广泛应用于遥感图像分析、目标检测等领域。
RSAR is a rotated SAR object detection dataset jointly constructed by Nankai University, Shanghai AI Laboratory and Shenzhen Futian NKIARI, and it is currently the largest multi-class dataset in this field. This dataset contains 95,842 SAR images and 183,534 annotated instances, covering six typical SAR object categories. The dataset construction process generates pseudo rotated bounding box labels via weakly-supervised models, followed by manual calibration to ensure the accuracy of annotations. The RSAR dataset aims to address the shortage of large-scale datasets in the field of rotated SAR object detection, improve the accuracy of model angle prediction, and is widely applied in remote sensing image analysis, object detection and other related fields.

- 1RSAR: Restricted State Angle Resolver and Rotated SAR Benchmark南开大学, 上海人工智能实验室, 深圳福田NKIARI · 2025年



