BEV-based 3D object detectors have gained recognition for their rapid processing capabilities, making them well-suited for on-device applications. In this research paper, we introduce SS-Pillar, a novel and efficient method for 3D object detection that offers superior quality results. Compared with the previous methods, we introduce the SS module which is the auxiliary network based on fine-grained pillars to learn the shape of objects. We also design a shape complete module to address the issue of far-distance missing region affected by the occlusion and sparse point clouds. Our model achieves real-time performance (28.29FPS) on NVIDIA Tesla V100 GPU significantly and outperforms competitive baselines on KITTI and nuScenes datasets.