Different axis convention for kikuchipy and orix

[htrellin]

Hi,

I have a question regarding the navigation shape convention when moving between kikuchipy and orix. It seems like they use two different conventions for x and y. Kikuchipy follows the convention (x, y) = (rows, columns) whereas Orix uses (x, y) = (columns, rows). Is there a reason for this difference?

import kikuchipy as kp
from orix import io

s = io.load("C:/EBSD data/DI data/Ni/di_results_7/di_ref_results_ni.ang")
s_2 = kp.load("C:/EBSD data/DI data/Ni/Pattern_19_67_14_24.h5")
print(f"orix: {s.shape}")
print(f"kikuchipy: {s_2.axes_manager.navigation_shape}")

which gives the results

orix: (10, 48)
kikuchipy: (48, 10)

[hakonanes]

It's the other way round, orix uses (n rows, n columns), which is in line with how NumPy represents a 2D array, while kikuchipy's signals extend HyperSpy's Signal2D class, which in all its representations use (n columns, n rows). This distinction is very important to be aware of! HyperSpy's rationale for this choice is explained in their documentation.

You can see the difference when comparing s_2.axes_manager.navigation_shape and s_2.data.shape.

[hakonanes]

Whenever the axes_manager navigation_shape or signal_shape is used in combination with orix or arrays, it is important to reverse the axes using for example navigation_shape[::-1]. This is done many places in kikuchipy's documentation, e.g. in cell 12 in the VBSE tutorial: https://kikuchipy.org/en/stable/tutorials/virtual_backscatter_electron_imaging.html#Generate-many-virtual-images.

[htrellin]

Thanks for the quick reply, I was not aware of this difference, so it helps a lot!

[hakonanes]

The complexity of going from 1D to 2D...

[hakonanes]

I took the liberty of transfering our discussion to the Q&A category, since you asked a question, and marking my comment as the answer. Hope this is OK.