Abstract

The previous point cloud compression methods only consider reducing the amount of data. However, in applications such as autonomous driving, the compression methods not only require smooth transmission, but also improve the efficiency of downstream tasks. To this end, we propose a task-driven sampling network based on graph convolution to achieve point cloud compression and recovery. First, we present a task-driven downsampling network based on graph convolution to compress the point cloud. Then, we present an upsampling network based on graph convolution to enhance and recover the point cloud. In order to optimize the compressed point cloud for task, we add the task loss to loss function for end-to-end training. Experiments for point cloud classification task on ModelNet40 dataset show that the compressed point cloud obtained through our network can achieve higher classification accuracy compared to other similar methods, and the reconstructed point cloud can further improve classification accuracy.