The dataset is/was available on https://www.datasciencechallenge.org/challenges/1/safe-passage/ . preprocess.py lets you transform the 2000x2000 images into 250x250 images and a CSV file with all the objects annotations. The dataset contains only the position of the center of the objects (no bounding boxes). A bounding box is generated. It's just a square centered on the provided position (x,y). The size of the square varies depending on the type of vehicle. We're using 8 object classes: Motorcycle, Light short rear, Light long rear, Dark short rear, Dark long rear, Red short rear, Red long rear, Light van. Other types of vehicles are ignored.