This is a large-scale object detection dataset with annotations. The dataset contains 365 categories of target objects, 2 million images, and 30 million bounding boxes, which are manually annotated through a carefully designed three-step annotation pipeline.

This dataset is the largest object detection dataset to date (with complete annotations), establishing a more challenging benchmark for the community. It can also serve as a good feature learning dataset for positioning-sensitive research such as object detection and semantic segmentation.