Reading image from array outputs invalid shape error

Question

I am very new at working with images in Python. I have an assignment which requires me to perform some preprocessing in images, which are given to me from an NPZ file. I have, then, split the data into train, validation and testing, and this is how it looks:

train_set = PathMNIST(mode="train") #89996
val_set = PathMNIST(mode="val") #10004
test_set = PathMNIST(mode="test") #7180

print(len(train_set))
print(len(val_set))
print(len(test_set))

And each item in the sets looks like this:

(array([220, 208, 227, ..., 222, 209, 228], dtype=uint8), 0)

So, as I understand it, it is a tuple, in which the first position represents the image, and the second position represents a label. What I am trying to do is to read the array in the first position as an image so I can apply some pre processings to it. However, when I try to plot it I get an Invalid shape error:

plt.figure(figsize=(10, 10))
for i, (images, labels) in enumerate(train_set):
  ax = plt.subplot(3, 3, i + 1)
  plt.imshow(images)
  plt.title(np.argmax(labels[i]))
  plt.axis("off")

#output: TypeError: Invalid shape (2352,) for image data

So I guess my question is: how do I read that array as an image and treat it as such? I looked at some questions and topics about it, but I cannot make it work for my scenario.

You would have to reshape your array before plotting it. Do you know the predefined shape of each image? — Jeru Luke
– Jeru Luke, Commented Jun 4, 2022 at 17:26
mnist data is usually given flattened to (784,) while the original images are (28, 28). you have three times as much data, so you seem to have color information too. — Christoph Rackwitz
– Christoph Rackwitz, Commented Jun 4, 2022 at 18:47
@Pelicer You could reshape your array using numpy: b = np.reshape(a, (28, 28,3)), but you would need to check the channel order — Jeru Luke
– Jeru Luke, Commented Jun 4, 2022 at 19:12
@Pelicer Every image has 3 channels depicting R,G and B. Each of these channels is an array of size (28*28), so that's why the shape: (28 * 28 *3). I do not know the order in which to place the channels, whether it should be RGB or BGR. — Jeru Luke
– Jeru Luke, Commented Jun 4, 2022 at 19:22

Jeru Luke · Accepted Answer · 2022-06-04 19:38:55Z

1

Your array is in 1D of size 2352.

As Christoph mentioned, every data point in the MNIST data set is an RGB image of size (28 * 28 * 3). And each of them have been flattened out to 1D array of size 2352.

We can reshape it using numpy:

b = np.reshape(a, (28, 28,3))

answered Jun 4, 2022 at 19:38

Jeru Luke

21.4k13 gold badges85 silver badges91 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Reading image from array outputs invalid shape error

1 Answer 1

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related