Darknet/YOLO People-R-People Weights

The Darknet/YOLO "People-R-People" weights are trained using a combination of parts from several different datasets. It uses the following:

• CrowdHuman, "Detecting Human in a Crowd"
• MS COCO, images without people are used to provide negative samples
• MS COCO, subset of class #0 ("person") are used to find 1, 2, or 3 people in an image
• Scraper, subset of safe-for-work images are used to find 1 or 2 people in an image

There are 2 classes defined:

1. class #0 is person
2. class #1 is head

The necessary files can be downloaded from here: download/.

Example Darknet/YOLO command to load the weights and process all JPEG images in a folder:

The People-R-People weights were trained using Darknet/YOLO.