Building an object detection model