I am trying to build a car type detector (sedan, van, etc) using a state of the art object detection algorithm. For that, I want to train with my own images of a parking lot.