Voxel51/GQA-Scene-Graph

GQA-35k是一个基于FiftyOne的数据集，包含35,000个样本。该数据集是GQA（Visual Reasoning in the Real World）数据集的一个子集，专注于视觉问答任务，但仅包含场景图注释，不包含问题。每个样本都包含详细的场景图，描述了图像中的对象、属性和关系。场景图基于Visual Genome场景图的清洁版本，提供了图像元数据、对象字典（包含对象名称、边界框坐标、属性和关系）以及关系三元组（指定谓词和目标对象ID）。数据集由Drew Hudson和Christopher Manning策划，由Voxel51的Harpreet Sahota共享，语言为英语，注释部分采用CC BY 4.0许可，图像可能具有不同的许可条款。

The GQA-35k dataset is a subset of the larger GQA (Visual Reasoning in the Real World) dataset, containing 35,000 samples focused on object detection. It includes scene graph annotations for each image, describing the objects, attributes, and relations present. The dataset is provided in a format compatible with FiftyOne, a tool for dataset management and visualization. The README also includes instructions for installation and usage of the dataset, as well as detailed descriptions of the dataset structure and annotations.