3dpcc/pace

PACE is an inference toolkit for learned LiDAR point-cloud compression. Its main workflow reads dataset-specific point clouds, quantizes and organizes them as octrees, predicts occupancy probabilities with a learned context model, and entropy-codes them into a bitstream; decoding reconstructs octree symbols, dequantizes coordinates, and writes a point cloud. Encoding supports parallel, autoregressive, and configurable multi-stage prediction. The README also documents bitrate and distortion evaluation. The excerpts show the decode path and parts of encoding; links for unsampled evaluation and encoding internals are included only where documented.