场景图生成SGG的效果评价指标

Scene Graph Detection (SGDet): given images as inputs, predict instance locations, categories, and relationships.场景图检测：输入图片，预测物体的位置，类别和物体与物体之间的关系
• Scene Graph Classification (SGCls): given images and instances locations, predict instance categories and relationships. 场景图分类：输入图片和物体的位置信息，预测物体类别以及关系。
• Predicate Classification (PredCls): given images, instance locations, and categories, predict relationships. 谓词分类：输入图片，物体位置信息与类别，预测物体之间的关系。
• Predicate detection (PredDet): given images, instance locations, categories, and relationship connections, predict relationship labels.
等等