Pixel and feature level based domain adaptation for object detection in autonomous driving