Remote Sensing Change Caption (RSCC)
收藏资源简介:
RSCC数据集是一个大规模的基准数据集,包含62,315对灾前灾后图像(涵盖地震、洪水、野火等事件),每对图像都配有人类风格的详细变化描述。该数据集填补了现有遥感数据集缺乏时间序列图像对和详细文本注释的空白,通过连接遥感数据中的时间和语义鸿沟,RSCC使视觉语言模型能够进行稳健的训练和评估,从而更好地理解灾害事件。数据集的构建过程包括数据来源、属性提取、提示构建、QvQ-Max推理、后修正和人工验证等步骤。RSCC数据集的应用领域包括灾害监测、分析和响应,旨在解决灾害事件中动态影响随时间变化的问题。
The RSCC dataset is a large-scale benchmark dataset containing 62,315 pairs of pre-disaster and post-disaster remote sensing images covering events such as earthquakes, floods, and wildfires. Each pair of images is accompanied by detailed human-written change descriptions. This dataset fills the gap in existing remote sensing datasets that lack paired time-series images and detailed textual annotations. By bridging the temporal and semantic gaps in remote sensing data, the RSCC dataset enables robust training and evaluation of vision-language models, allowing them to better understand disaster events. The construction process of the RSCC dataset includes steps such as data sourcing, attribute extraction, prompt engineering, QvQ-Max inference, post-correction, and manual verification. The application scenarios of the RSCC dataset cover disaster monitoring, analysis, and response, aiming to address the issue of dynamically changing impacts of disaster events over time.
- 1RSCC: A Large-Scale Remote Sensing Change Caption Dataset for Disaster Events浙江大学 · 2025年



