ENH: 3d examples for filters (e.g. `gaussian_filter`)

Is your feature request related to a problem? Please describe.

Right now, there are only 2D examples for Multidimensional filters, e.g. the Gaussian filter: https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.gaussian_filter.html

Describe the solution you’d like.

Provide also 3D examples. Toy examples in 3D will help to make the filter parameters easier to understand. I had the idea to use 3D voxel for that.

Describe alternatives you’ve considered.

No response

Additional context (e.g. screenshots, GIFs)

Either as animation:

https://user-images.githubusercontent.com/44469195/141647703-cf3c597c-e404-4fef-b746-495d59146685.mp4

Or as images

This is only a quick code draft for reproduction (not very pretty):

import matplotlib.pyplot as plt
import numpy as np
import matplotlib as mpl
from scipy.ndimage import gaussian_filter
%config InlineBackend.figure_format = 'svg'

size= 10
x, y, z = np.indices((size*2, size*2, size*2))

o1,o2,o3 =(0,0,0)
cube1 = (o1 <= x) & (x < o1+size) & (o2 <= y) & (y < o2+size) &  (o3 <= z) & (z < o3+size) 

o1,o2,o3 =(size,0,0)
cube2 = (o1 <= x) & (x < o1+size) & (o2 <= y) & (y < o2+size) &  (o3 <= z) & (z < o3+size) 

o1,o2,o3 =(0,size,0)
cube3 = (o1 <= x) & (x < o1+size) & (o2 <= y) & (y < o2+size) &  (o3 <= z) & (z < o3+size) 

o1,o2,o3 =(size,size,0)
cube4 = (o1 <= x) & (x < o1+size) & (o2 <= y) & (y < o2+size) &  (o3 <= z) & (z < o3+size) 

o1,o2,o3 =(0,0,size)
cube5 = (o1 <= x) & (x < o1+size) & (o2 <= y) & (y < o2+size) &  (o3 <= z) & (z < o3+size) 

o1,o2,o3 =(size,0,size)
cube6 = (o1 <= x) & (x < o1+size) & (o2 <= y) & (y < o2+size) &  (o3 <= z) & (z < o3+size) 

o1,o2,o3 =(0,size,size)
cube7 = (o1 <= x) & (x < o1+size) & (o2 <= y) & (y < o2+size) &  (o3 <= z) & (z < o3+size) 

o1,o2,o3 =(size,size,size)
cube8 = (o1 <= x) & (x < o1+size) & (o2 <= y) & (y < o2+size) &  (o3 <= z) & (z < o3+size) 


voxelarray = np.zeros(cube1.shape)

voxelarray[cube1]=1 * 255/8
voxelarray[cube2]=2 * 255/8
voxelarray[cube3]=3 * 255/8
voxelarray[cube4]=4 * 255/8
voxelarray[cube5]=5 * 255/8
voxelarray[cube6]=6 * 255/8
voxelarray[cube7]=7 * 255/8
voxelarray[cube8]=8 * 255/8
voxelarray = np.uint8(voxelarray)

cmap = plt.get_cmap('magma')    
cols256=[]
for i in range(cmap.N):
    rgb = cmap(i)[:3] 
    cols256.append(mpl.colors.rgb2hex(rgb))

def get_color(x):
    return cols256[x]
get_color = np.vectorize(get_color)

colors = get_color(voxelarray)


def plot_voxels(varray,title):
    colors = get_color(varray)
    global ax
    ax = plt.figure().add_subplot(projection='3d')
    plt.axis('off')
    ax.voxels(varray, facecolors=colors, edgecolor="#000000",linewidth=0.1)
    ax.set_title(title) 
plot_voxels(voxelarray,"Original")
voxelarray2=gaussian_filter(voxelarray, sigma=1)
plot_voxels(voxelarray2,"gaussian_filter(voxelarray, sigma=1)")
voxelarray2=gaussian_filter(voxelarray, sigma=3)
plot_voxels(voxelarray2,"gaussian_filter(voxelarray, sigma=3)")

[kolibril13]

That would be possible. But should all the code be displayed? It is about 70 lines in total, but only these 6 lines are actually important to understand the gaussian_filter:

from scipy.ndimage import gaussian_filter
plot_voxels(voxelarray,"Original")
voxelarray2=gaussian_filter(voxelarray, sigma=1)
plot_voxels(voxelarray2,"gaussian_filter(voxelarray, sigma=1)")
voxelarray2=gaussian_filter(voxelarray, sigma=3)
plot_voxels(voxelarray2,"gaussian_filter(voxelarray, sigma=3)")

[tupui]

Sounds good!

I can refactor the code a bit and then make a pull request.

@tupui : Would you be willing to make some fine-tuning of the formatting and commit them directly to my soon coming pull request branch, so that my example is embedded nicely in the style of the docs?

Sure of course! Thanks for making a PR 😃