Numpy原位转置

Question

I have a large numpy ndarray shaped (H, W, 3)

I need to do this

new_arr = arr[:, :, [2,1,0]]

I timed it and it takes a few ms. And my understanding is that this is a copy. Is there an place operation that is faster?

Answer 1

I recently made a library for performing in-place transpositions for 2d, 3d, and 4d arrays. In place transposition is faster than numpy for square matrices. Otherwise, it is more memory efficient, but slower.

https://github.com/seung-lab/fastremap