The NWPU-Crowd dataset contains 5,109 common crowd images.

The images are from common search engines such as Google, Baidu, Bing, Sougou, and image material websites such as pixbay, pxhere, and wallhere. They are retrieved and collected using common keywords such as "crowd, crowded, Spring Festival travel, military training", aiming to include diverse crowd scenes. They are annotated in the form of head points and head frames, involving a total of 2,133,375 heads.

Compared with other datasets, this dataset contains various lighting scenes, and the number of people in a single image ranges from 0 to 20,033.