Self-Attention Based Context-Aware 3D Object Detection