3D object detection from point cloud is an important aspect of environmental perception in intelligent systems such as autonomous driving systems and robot systems. However, efficient 3D feature extraction and accurate object localization is challenging for current algorithms. In this paper, we introduce a new single-shot refinement neural network for fast and accurate 3D object detection. Firstly, we simplify the 3D feature extraction network and use single-shot object detector to increase processing speed. Secondly, we exploit self-attention mechanism in main object detection branch to improve object feature representation. Thirdly, an object refinement branch is introduced to produce a finer regression of objects upon the primary estimation from the main detection branch. Both modifications lead to further improvements in performance without additional computational cost. Our approach is tested on KITTI 3D Car detection benchmark and achieves good results in the validation set. The running speed is around 40 frame per second.