CODEWITHSUNDEEP
pythonarrayspython-3.xnumpyresize
Onur Dasdemir
Onur Dasdemir

Reputation: 67

How to upsample an array to arbitrary sizes?

I am trying to resize an array to a larger size in Python by repeating each element proportionally to the new size. However, I want to be able to resize to arbitrary sizes.

I know that I can do it with numpy.repeat if for example I have to double the size but lets say I want to convert an array of size (180,150) to (300,250). I know there is not a perfect way to do this but I am looking for the most efficient (minimum loss of information) method!

So far, I was converting the array to an image and resize it accordingly, then convert it to an array again. However, it seems that I cannot convert all types of data to image so I need a general way to do this.

For example, lets say I have an input array of size (2,2):

input_array=np.array([[1,2],[3,4]])

If I want to convert it to a (3,3) array, output may be like:

output_array=np.array([[1,1,2],[1,1,2],[3,3,4]])

Like I said before, I just don't want to tile or fill with zeros, I want to expand the size by repeating some of the elements.

Upvotes: 1

Views: 1895

Answers (2)

tstanisl
tstanisl

Reputation: 14167

If you look for pure numpy solution then you can try to use fancy indexing:

outshape = 3,3
rows = np.linspace(0, input_array.shape[0], endpoint=False, num=outshape[0], dtype=int)
cols = np.linspace(0, input_array.shape[1], endpoint=False, num=outshape[1], dtype=int)
# Extract result using compute indices
output_array=input_array[rows,:][:,cols]

Upvotes: 3

sentence
sentence

Reputation: 8933

Without a clear idea about the final result you would like to achieve, your question opens multiple paths and solutions. Just to name a few:

  1. Using numpy.resize:
import numpy as np

input_array=np.array([[1.,2],[3,4]])

np.resize(input_array, (3,3))

you get:

array([[1., 2., 3.],
       [4., 1., 2.],
       [3., 4., 1.]])
  1. Using cv2.resize:
import cv2
import numpy as np

input_array=np.array([[1.,2],[3,4]])

cv2.resize(input_array,
           (3,3),
           interpolation=cv2.INTER_NEAREST)

you get:

array([[1., 1., 2.],
       [1., 1., 2.],
       [3., 3., 4.]])

Depending on your objective, you can use different interpolation methods.

Upvotes: 3

Related Questions