VAW

该数据集由VGPhraseCut和GQA构建而成，经过清洗后，包含了620个描述物体颜色、材质、形状、大小、纹理和动作的属性。在无边界框设置的零样本属性分类任务上进行了评估。规模上，该数据集涵盖了620个描述各种物体特征属性的属性，任务包括属性分类和属性定位。

This dataset is constructed from VGPhraseCut and GQA. After data cleaning, it contains 620 attributes describing the color, material, shape, size, texture and action of objects. It is evaluated on the bounding box-free zero-shot attribute classification task. In terms of scale, the dataset covers 620 attributes characterizing various object features, and the tasks include attribute classification and attribute localization.