RSVLM-QA
收藏资源简介:
RSVLM-QA是一个针对遥感视觉语言模型问答的大规模、多源、丰富注释的VQA数据集。它由WHU、LoveDA、INRIA和iSAID四个遥感数据集整合而成,并采用了一种创新的LLM驱动(GPT-4.1)的注释生成流水线。该数据集包含13,820张图像和162,373个VQA对,具有广泛的注释和多样化的题型。RSVLM-QA旨在解决遥感图像中复杂的解释和问答问题,为地球观测数据的解释提供支持。
RSVLM-QA is a large-scale, multi-source, richly annotated Visual Question Answering (VQA) dataset designed for question answering tasks of remote sensing vision-language models. It integrates four remote sensing datasets including WHU, LoveDA, INRIA, and iSAID, and employs an innovative LLM-driven (GPT-4.1) annotation generation pipeline. The dataset consists of 13,820 images and 162,373 VQA pairs, featuring comprehensive annotations and diverse question types. RSVLM-QA aims to address complex interpretation and question answering challenges in remote sensing imagery, providing support for the interpretation of Earth observation data.

- 1RSVLM-QA: A Benchmark Dataset for Remote Sensing Vision Language Model-based Question AnsweringUniversity of Technology Sydney · 2025年



