A Lightweight Model for 3D Point Cloud Object Detection

With the rapid development of deep learning, more and complex models are applied to 3D point cloud object detection improve accuracy. In general, model, better performance greater computational resource consumption it has. However, incompatible for deployment on edge devices with restricted memory, so accurate efficient processing is necessary. Recently, a lightweight model design has been proposed as one type effective compression that aims network computing methods. this paper, architecture proposed. The core innovation proposal consists sparse convolution layer module (LW-Sconv module) knowledge distillation loss. Firstly, in LW-Sconv module, factorized group standard layer. As basic component greatly reduces complexity network. Then, loss used guide training paper further Finally, extensive experiments performed verify algorithm paper. Compared baseline can reduce FLOPs parameters by 3.7 times 7.9 times, respectively. trained achieves comparable accuracy baseline. Experiments show method while ensuring