Parallel crop point cloud by bounding box list

[HowcanoeWang]

Hey,

Currently, I have thousands of small regions want to crop from large point cloud (50 millions of points), each small region save to an individual pcd file, and I am writing a for loop to implement this:

o3d_pcd = my_function_to_prepare_pcd_data()

for bbox in bbox_list:   # bbox is a polygon on XY plane, len(bbox_list) > 1000
    # make bounding box from bbox
    o3d_boundary = o3d.visualization.SelectionPolygonVolume()
    o3d_boundary.orthogonal_axis = "Z"
    o3d_boundary.bounding_polygon = o3d.utility.Vector3dVector(bbox)
    o3d_boundary.axis_max = z_max
    o3d_boundary.axis_min = z_min

    # crop by previous bounding box
    pcd_crop = o3d_boundry.crop_point_cloud(o3d_pcd)
    o3d.io.write_point_cloud(bbox_name, pcd_crop)

However, the previous loop may take over an hour to operate (round 10s per loop) one point cloud, this is quite time-costly (I have over 100 such pcd to process), will it possible to crop the point cloud faster by those small regions? Either crop at once and then split the cropping results or by parallel computing? (Run on a PC with 32 core CPU and 128GB RAM)