The paper introduces a novel method called SECOND (Sparsely Embedded CONvolutional Detection) for 3D object detection using LiDAR data. The method addresses the limitations of existing approaches, such as slow inference speed and poor orientation estimation performance. SECOND incorporates several improvements to the convolutional network architecture, including sparse convolution for efficient training and inference, a new angle loss regression approach for better orientation estimation, and a data augmentation technique to enhance convergence speed and performance. The proposed method achieves state-of-the-art results on the KITTI 3D object detection benchmarks while maintaining fast inference speeds. The key contributions of the work are the application of sparse convolution in LiDAR-based object detection, an improved sparse convolution method, a novel angle loss regression approach, and a novel data augmentation method. The experiments demonstrate that the proposed method outperforms other state-of-the-art approaches and can run in real-time.The paper introduces a novel method called SECOND (Sparsely Embedded CONvolutional Detection) for 3D object detection using LiDAR data. The method addresses the limitations of existing approaches, such as slow inference speed and poor orientation estimation performance. SECOND incorporates several improvements to the convolutional network architecture, including sparse convolution for efficient training and inference, a new angle loss regression approach for better orientation estimation, and a data augmentation technique to enhance convergence speed and performance. The proposed method achieves state-of-the-art results on the KITTI 3D object detection benchmarks while maintaining fast inference speeds. The key contributions of the work are the application of sparse convolution in LiDAR-based object detection, an improved sparse convolution method, a novel angle loss regression approach, and a novel data augmentation method. The experiments demonstrate that the proposed method outperforms other state-of-the-art approaches and can run in real-time.