resize and pad numpy array with additional dimensions-CodePudding

I have an np array with shape (100,28,28)

I would like to turn it to (100,32,32,3)

by padding with zero's and adding dimensions

I tried np.newaxis which got me as far as (100,28,28,1)

CodePudding user response：

You can try the resize function on your array. This is the code I tried.

arr = np.ones((100,28,28))
arr.resize((100, 32, 32, 3))
print(arr)

The newly added items are all 0s.

CodePudding user response：

There is not enough information about how you want the padding to be done, so I'll suppose an image processing setup:

Padding on both edges of the array
Array must be replicated along the newly added dimension

Setup

import numpy as np

N, H, W, C = 100, 28, 28, 3
x = np.ones((N, H, W))

1. Padding

# (before, after)
padN = (0, 0)
padH = (2, 2)
padW = (2, 2)

x_pad = np.pad(x, (padN, padH, padW))

assert x_pad.shape == (N   sum(padN), H   sum(padH), W   sum(padW))

2. Replicate along a new dimensions

x_newd = np.repeat(x_pad[..., None], repeats=C, axis=-1)

assert x_newd.shape == (*x_pad.shape, C)
assert np.all(x_newd[..., 0] == x_newd[..., 1])
assert np.all(x_newd[..., 1] == x_newd[..., 2])