Rotating 5D cube with NumPy

Question

I got 2x 5D cubes in the shape of [1, 4, 21, 302, 302] that I need to compare with each other. However the first is rotated with respect to the second. If I work with just these two images I can fix it by applying np.rot90(np.flipud(a)) to the left image, with a being that 302x302 image. How do I apply this to the entire cube without changing the other axes?

Answer 1

Rotating the image like you're describing it is simply transposing/swapping the two axes x and y. For a single image in a 2D array, you would simply do

img.T

And for a 5D tensor like yours where the image is contained in the last two axes, you would do

img.transpose(0, 1, 2, 4, 3)