Build a very large pedestrian detection data set

Summary

Pedestrian detection is a complex but commonly used task and is the prerequisite for tasks such as tracking and ReID. By constructing ultra-large-scale pedestrian detection data sets in multiple scenarios, the generalization of the pedestrian detection model can be greatly improved. I extracted pedestrian category data from multiple datasets and constructed a very large-scale pedestrian detection dataset. And convert it into labelme-marked data to facilitate subsequent modification.