bad performance with warp_polar in 3D

[patquem]

Description:

Hello,
As mentioned in the documentation, warp_polar can be used in 3D.
Meanwhile, the performance are bad* in comparison to classical 2D warp_polar applied n-times on an array of shape (p, q) (considering an original 3D array of shape (n, p, q)).
Can this be improved ?
Thanks,
Patrick

*+60 % in the example below with my PC

Way to reproduce:

import time
import numpy as np
import matplotlib.pyplot as plt
from skimage.transform import warp_polar

arr = np.random.random((1000, 256, 256))

t1 = time.process_time()
warp_polar(arr, radius=100, channel_axis=0)
time1 = time.process_time() - t1

t2 = time.process_time()
for arr_i in arr:
    warp_polar(arr_i, radius=100)
time2 = time.process_time() - t2

print(time1 / time2)

Traceback or output:

No response

Version information:

3.7.7 (tags/v3.7.7:d7c567b08f, Mar 10 2020, 10:41:24) [MSC v.1900 64 bit (AMD64)]
Windows-10-10.0.19041-SP0
scikit-image version: 0.19.3
numpy version: 1.21.6

[stefanv]

My guess is that the for loops uses a lot less memory; that can sometimes help with performance. It's only 500MB of memory for the image, but the transformed coordinates also have to be computed and stored, so that's probably another 1500MB.

Even faster would be to precompute the warping coordinates for one slice, and to then re-apply those same coordinates repeatedly for each next slice.

[github-actions]

Hey, there hasn't been any activity on this issue for more than 180 days. For now, we have marked it as "dormant" until there is some new activity. You are welcome to reach out to people by mentioning them here or on our forum if you need more feedback! If you think that this issue is no longer relevant, you may close it by yourself; otherwise, we may do it at some point (either way, it will be done manually). In any case, thank you for your contributions so far!