RS-4M
收藏资源简介:
RS-4M是一个大规模的光学遥感数据集,由国防科技大学等机构创建,包含400万张图像,旨在支持高效的掩码图像建模(MIM)训练。数据集涵盖丰富的细粒度遥感视觉任务,如目标级检测和像素级分割。创建过程中,数据集排除了多光谱和SAR数据,专注于光学图像,并通过随机裁剪高分辨率图像来处理大规模图像分割。RS-4M的应用领域广泛,包括生态监测、自然灾害管理等,旨在通过提供大规模、多样化的数据集,解决现有遥感数据集在规模和多样性上的不足,提升下游任务的性能。
RS-4M is a large-scale optical remote sensing dataset created by institutions such as the National University of Defense Technology, comprising 4 million images. It is designed to support efficient Masked Image Modeling (MIM) training. This dataset covers a diverse set of fine-grained remote sensing visual tasks, including object-level detection and pixel-level segmentation. During its development, the dataset excluded multispectral and Synthetic Aperture Radar (SAR) data, focusing exclusively on optical imagery, and adopted random cropping of high-resolution images to handle large-scale image segmentation tasks. RS-4M has a wide range of application scenarios, such as ecological monitoring and natural disaster management. It aims to address the shortcomings of existing remote sensing datasets in terms of scale and diversity by providing a large-scale and diverse dataset, thereby enhancing the performance of downstream tasks.




