PyTorch reshepe 4d to 2d, with "with saving 1 dim"

Asked Apr 14 '20 at 01:39

Active Apr 14 '20 at 12:29

Viewed 26 times

Have a tensor X:

X.shape = [2, 3, 4, 5]

This tensor represents pictures.
X[0] - number of pic.
X[1] -  Chanels
X[2],X[3] -  Height and Width.

I want to convert this tensor to shape [40,2] where second dim is 3 channels, and first all pixels in input data, which correspond channels in right order. try to use

reshape(-1,3)

but chanels are in incorrect order. How i can convert it? It is also interesting how i can to convert back, from [40,2] to [2,3,4,5]

SOLUTION

Find solution fo my case:

x.permute(1,0,2,3).reshape(3,-1).T

from this X:

to this X:

edited Apr 14 '20 at 12:29

asked Apr 14 '20 at 01:39

iGero

[40,2] or [40,3] for the second one? – ilke444 Apr 14 '20 at 06:27
3

Does this answer your question? [Tips or patterns for reshaping 4D/5D arrays, (videos to frames)](https://stackoverflow.com/questions/50502700/tips-or-patterns-for-reshaping-4d-5d-arrays-videos-to-frames) – ilke444 Apr 14 '20 at 06:30
What is the right order? – Mad Physicist Apr 14 '20 at 07:39
@like444 Yes. [40,2]. Fixed it. – iGero Apr 14 '20 at 12:20

PyTorch reshepe 4d to 2d, with "with saving 1 dim"

SOLUTION

0 Answers0