Description

Open Images Dataset V6是谷歌开源的一个强大的图像公开数据集，里面包含约 900 万张图像，600个类别。可用于图像分类、对象检测、视觉关系检测、实例分割和多模态图像描述。

These annotation files cover the 600 boxable object classes, and span the 1,743,042 training    images where we annotated bounding boxes, object segmentations, visual relationships, and    localized narratives; as well as the full validation (41,620 images) and test (125,436 images)    sets.