Numpy reshape RGB image to 1D array

Description of this Notebook

Quick Links

Quick Links: Code

Imagine you want to train a NN with:

• X: s x 3 x 32 x 32

Where:

• s: num samples
• 3 x 32 x 32: color img

Now to feed in the NN we want the shapes to be:

• X: s x (3072)

Do we lose information? The answers is not, we just reshape X.

The following takes 2 imgs, with dimensions 2x2:

import numpy as np

# Create example 3x2x2 images (two images)
original_images = np.array([[[[1, 2],
                             [3, 4]],
                            [[5, 6],
                             [7, 8]],
                            [[9, 10],
                             [11, 12]]],

                           [[[13, 14],
                             [15, 16]],
                            [[17, 18],
                             [19, 20]],
                            [[21, 22],
                             [23, 24]]]])

# Print the shape of the original images
print("Original Images Shape:", original_images.shape)

# Print the original images
print("Original Images:\n", original_images)

# Reshape the images into num_imgs x (3*2*2) matrices
num_imgs = original_images.shape[0]
reshaped_images = original_images.reshape(num_imgs, -1)

# Print the shape of the reshaped images
print("Reshaped Images Shape:", reshaped_images.shape)

# Print the reshaped images
print("Reshaped Images:\n", reshaped_images)

Original Images Shape: (2, 3, 2, 2)
Original Images:
 [[[[ 1  2]
   [ 3  4]]

  [[ 5  6]
   [ 7  8]]

  [[ 9 10]
   [11 12]]]


 [[[13 14]
   [15 16]]

  [[17 18]
   [19 20]]

  [[21 22]
   [23 24]]]]
Reshaped Images Shape: (2, 12)
Reshaped Images:
 [[ 1  2  3  4  5  6  7  8  9 10 11 12]
 [13 14 15 16 17 18 19 20 21 22 23 24]]