WayBob/Disaster_Recognition_RemoteSense_EN_CN_JA

xView2多语言灾难识别数据集是基于xBD (xView2)建筑损坏评估数据集重新格式化的，专为支持多语言的视觉语言模型(VLM)训练而设计。该数据集包含卫星图像与多语言对话注释的配对，支持英语、中文和日语三种语言。数据集结构包括训练集和测试集，每个图像对包含灾前和灾后卫星图像、分割掩码、建筑损坏标签和元数据。数据集覆盖了六种自然灾害类型：火山、洪水、风灾、地震、海啸和火灾。数据集适用于视觉语言模型训练、多语言AI系统、灾难评估、变化检测和人道主义AI应用等场景。

The xView2 Multi-Language Disaster Recognition Dataset is derived from the xBD (xView2) Building Damage Assessment Dataset and has been reformatted for Vision-Language Model (VLM) training with multi-language support. This dataset contains satellite imagery paired with multi-language conversational annotations, supporting three languages: English, Chinese, and Japanese. The dataset structure includes training and test sets, with each image pair consisting of pre-disaster and post-disaster satellite images, segmentation masks, building damage labels, and metadata. The dataset covers six types of natural disasters: volcano, flooding, wind damage, earthquake, tsunami, and fire. It is suitable for vision-language model training, multi-language AI systems, disaster assessment, change detection, and humanitarian AI applications.