HRVQA
收藏资源简介:
HRVQA是一个专为高分辨率航空图像设计的视觉问答数据集,由特文特大学创建。数据集包含53512张1024x1024像素的航空图像,涵盖多种场景和概念,如工业、港口、车站和体育设施。此外,数据集还包含了1070240个半自动生成的QA对,覆盖10种不同类型的问题,从常见的任务推理到特定属性的识别。创建过程中,采用了模板策略和对象检测模型来生成问题和答案。HRVQA旨在为计算机视觉和地球观测研究提供一个全面的基准,特别适用于场景理解和地理空间推理任务。
HRVQA is a visual question answering (VQA) dataset specifically designed for high-resolution aerial images, created by the University of Twente. The dataset contains 53,512 aerial images with a resolution of 1024×1024 pixels, covering diverse scenarios and concepts such as industrial areas, ports, stations, and sports facilities. Additionally, it includes 1,070,240 semi-automatically generated QA pairs that cover 10 distinct types of questions, ranging from common task-based reasoning to specific attribute recognition. During the dataset creation, template strategies and object detection models were adopted to generate the questions and answers. HRVQA aims to provide a comprehensive benchmark for computer vision and earth observation research, and is particularly applicable to scene understanding and geospatial reasoning tasks.

- 1HRVQA: A Visual Question Answering Benchmark for High-Resolution Aerial Images特文特大学 · 2023年



