Volumetric movies (3D video captured by depth sensors) generate massive datasets. Storing raw 3D coordinate frames results in unmanageable file sizes. "New" PointNet adaptations are used to compress these point clouds by learning global geometry features, allowing high-quality spatial movies to be tightly packed into standard MKV containers and streamed efficiently. 2D-to-3D Video Conversion
[1612.00593] PointNet: Deep Learning on Point Sets for 3D ... - arXiv mkv movies pointnet new
If you want to dig deeper into setting up this pipeline, please let me know: Volumetric movies (3D video captured by depth sensors)