Use list comprehension to move data from an array to another array using index of a 3rd array

Question

I was am trying to find more efficent and faster way of unpacking a list of objects into its component parts. The approach i have currently working is this:

points_fill_v2 = o3d.geometry.VoxelGrid.create_from_point_cloud(points,
                                                            voxel_size=vox)
scrap = points_fill_v2 .get_voxels()
resize3d = np.ones((3, 64, 64, 64))

for idx in range(len(scrap)):
    resize3d[0, scrap[idx].grid_index[0], scrap[idx].grid_index[1], scrap[idx].grid_index[2]] = scrap[idx].color[0]
    resize3d[1, scrap[idx].grid_index[0], scrap[idx].grid_index[1], scrap[idx].grid_index[2]] = scrap[idx].color[1]
    resize3d[2, scrap[idx].grid_index[0], scrap[idx].grid_index[1], scrap[idx].grid_index[2]] = scrap[idx].color[2]

scrap is a list and looks like this:

[Voxel with grid_index: (7, 27, 30), color: (0, 0, 1), Voxel with grid_index: (32, 9, 8), color: (0, 0, 0), ......] 

It has 2 properties; grid index and color, both are numpy 1,3 arrays.

The above code converts a point cloud into a voxel dataset using open3d (I have to use open3d), and then converts that into a RGB 3d numpy array.

I can split out the main list into 2 other lists (and then arrays) using:

grid = [scrap[idx].grid_index for idx in range(len(scrap))]
colour = [scrap[idx].color for idx in range(len(scrap))]

How do I transfer the data in colour into resize_3d efficiently using the index stored in grid?

Answer 1

This refactoring should give you a slight speed increase, and makes your code a lot cleaner:

from itertools import product
 
points_fill_v2 = o3d.geometry.VoxelGrid.create_from_point_cloud(points, voxel_size=vox)
scrap = points_fill_v2.get_voxels()
resize3d = np.ones((3, 64, 64, 64))

for (idx, scrap_elem), j in product(enumerate(scrap), range(3)):
    resize3d[(j, *scrap_elem.grid_index[:3])] = scrap_elem.color[j]

Things I've done here:

• I've used enumerate(<sequence>) instead of range(len(<sequence>)).
• I've used itertools.product to reduce three lines of code into one.
• I've used slicing, combined with star-unpacking, to simplify some of your list-indexing code, which also has the effect of reducing the number of dot-lookups per iteration.