At this point PointCloudRasterSources uses PDAL JNI which uses PDAL Pipeline to read entwine tiles. However, this can be slow and not parallelizable.
To resolve this task it is possible to use Entwine Reader directly.
The cost is also that we won't be able to perform any of the Pipeline operations on such points. So probably it would also require implementing triangulation ourselves on the read data. However, if that could be done through the Pipeline, probably we can re-use its PDAL implementation somehow.
If this approach would be slow it looks like it would be to do nothing better in terms of the IO performance. See the current benchmarks module.
Summary:
[ ] Expose Enwine Reader through JNI bindings (probably it deserves a separate library / module in the PDAL/java repository)
[ ] Use PDAL triangulation to compute mesh (this would be a C call)
[ ] Rasterize the mesh it (this would be also on the C call side)
At this point PointCloudRasterSources uses PDAL JNI which uses PDAL Pipeline to read entwine tiles. However, this can be slow and not parallelizable.
To resolve this task it is possible to use Entwine Reader directly.
The cost is also that we won't be able to perform any of the Pipeline operations on such points. So probably it would also require implementing triangulation ourselves on the read data. However, if that could be done through the Pipeline, probably we can re-use its PDAL implementation somehow.
If this approach would be slow it looks like it would be to do nothing better in terms of the IO performance. See the current benchmarks module.
Summary:
Connects https://github.com/geotrellis/geotrellis-pointcloud/issues/74