Performance of threading read_point_cloud

Frequently I need to load multiple pcd files, which takes some time when loading sequentially.

So I tried using threading to execute several o3d.io.read_point_cloud calls concurrently.

But the performance was not any better. Plenty of IO bandwidth on the SSD, and plenty of spare CPU cores too.

It’s probably a Python GIL (Global Interpreter Lock) issue.

Anyone know if there’s a simple fix to release the GIL in pybind?

Hi,

I got similar problem. The parallel processing is one thing. I also find open3d.io.read_pointcloud automatically returns “Format=pcd, Extension=auto” to console after the successful loading of each point cloud. It also reduces the processing speed. Is there any way we can remove this print info from console?