ReCon1M
收藏资源简介:
ReCon1M是由中国科学院航空信息研究所创建的大型遥感图像关系理解数据集。该数据集基于Fair1M数据集,包含21,392张图像,标注了859,751个对象边界框和1,149,342个关系三元组,覆盖60个对象类别和64个关系类别。数据集的创建过程涉及专家指导下的新对象标注和统一标注标准的制定。ReCon1M旨在推动遥感图像中对象关系理解的研究,特别是在场景图生成任务中,为模型提供丰富的训练资源和评估基准,以解决现有模型在理解复杂对象关系方面的不足。
ReCon1M is a large-scale remote sensing image relationship understanding dataset developed by the Aviation Information Research Institute of the Chinese Academy of Sciences. Built upon the Fair1M dataset, it contains 21,392 images, annotated with 859,751 object bounding boxes and 1,149,342 relationship triplets, covering 60 object categories and 64 relationship categories. The development of ReCon1M involved the formulation of unified annotation standards and the creation of new object annotations under expert supervision. ReCon1M aims to advance research on object relationship understanding in remote sensing images, particularly for the scene graph generation task, by providing abundant training resources and evaluation benchmarks for models to address the shortcomings of existing models in comprehending complex object relationships.

- 1ReCon1M:A Large-scale Benchmark Dataset for Relation Comprehension in Remote Sensing Imagery中国科学院航空信息研究所 · 2024年



