Benchmark of compressing sequential point clouds.
To run the benchmark, you need a Python environment with the following packages: numpy, DracoPy, pcl.py, laspy[lazrs].
You also need a 100-frame sequence of point cloud data.
The tested data used here are generated with the CARLA simulator. Each point in the cloud has a structure of (x, y, z, intensity), all represented by 32-bit floats.
| Compression Time | Compressed Size | Compression Ratio | |
|---|---|---|---|
| tar.xz (baseline) | 12.595 | 16.25213 | 17.21638 |
| numpy npz | 4.254 | 26.37437 | 27.93917 |
| las + tar.xz | 4.877 | 10.0848 | 10.68314 |
| draco (q=14) + tar.xz | 6.004 | 8.701447 | 9.217706 |
| draco (q=10) + tar.xz | 4.314 | 5.032021 | 5.330572 |
| pcl + tar.xz | 19.434 | 23.67338 | 25.07793 |
| las aggregated | 0.405 | 9.952751 | 10.54325 |
| draco (q=14) aggregated | 37.712 | 8.812826 | 9.335694 |
| draco (q=10) aggregated | 38.127 | 5.535102 | 5.863501 |
| las aggregated (w/pose) | 0.378 | 9.86227 | 10.4474 |
| draco (q=14) aggregated w/pose | 38.276 | 7.205489 | 7.632993 |
| draco (q=10) aggregated w/pose | 38.966 | 4.245287 | 4.497161 |