Towards Generalized 3D Object Detection for Autonomous Driving: Enhancing Data Efficiency in Novel Domains