Lidar Range Image Compression with Deep Delta Encoding

Lidars are widely used in applications such as autonomous driving and augmented reality. However, the large volume of data produced by lidars can lead to high cost in data storage and transmission. Besides rising standards in point cloud compression from the MPEG (G-PCC and V-PCC), recent works also explore using deep networks to improve the compression rates. However, most prior work focus on the generic point cloud representation, neglecting the spatial patterns of the points from lidar range images. In this work, we leverage the range image representation and propose a novel deep delta encoding model to compress lidar data. Our deep model takes in local range image patches and predicts the next pixel value in a raster-scanning manner. The residuals between the prediction and the original value can be entropy encoded to achieve lossless compression under certain quantization rates. Evaluated on the Waymo Open Dataset and KITTI, our method demonstrates significant improvement compared to widely algorithms as well as recent deep methods based on the point cloud representation, in both the point cloud reconstruction quality and the downstream perception model performance.