Python matplotlib, invalid shape for image data

Question

Currently I have this code to show three images:

imshow(image1, title='1')
imshow(image2, title='2')
imshow(image3, title='3')

And it works fine. But I am trying to put them all three in a row instead of column.

Here is the code I have tried:

f = plt.figure()
f.add_subplot(1,3,1)
plt.imshow(image1)
f.add_subplot(1,3,2)
plt.imshow(image2)
f.add_subplot(1,3,3)
plt.imshow(image3)

It throws

TypeError: can't convert CUDA tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.

If I do

f = plt.figure()
f.add_subplot(1,3,1)
plt.imshow(image1.cpu())
f.add_subplot(1,3,2)
plt.imshow(image2.cpu())
f.add_subplot(1,3,3)
plt.imshow(image3.cpu())

It throws

TypeError: Invalid shape (1, 3, 128, 128) for image data

How should I fix this or is there an easier way to implement it?

Use Matplotlib's subplots function, specifying the number of rows you want through the ncols argument. — Lith
– Lith, Commented Apr 28, 2020 at 13:35
@Lith I have tried with this code: fig, axs = plt.subplots(nrows=1, ncols=3) axs[0].imshow(image1.cpu()) axs[1].imshow(image2.cpu()) axs[2].imshow(image3.cpu()) and still getting TypeError: Invalid shape (1, 3, 128, 128) for image data — Hekes Pekes
– Hekes Pekes, Commented Apr 28, 2020 at 14:11
I don't know about Pytorch, but it seems that the cpu() method is transforming the array image with (I suppose) dimensions (128,128) into another of dimensions (1,3,128,128), which is invalid for the imshow function. The argument of this function must be a two dimensional array (or 3-dimensional if you are working with RGB data), representing the data values of the pixels of the image. — Lith
– Lith, Commented Apr 28, 2020 at 14:53

A. Maman · Accepted Answer · 2020-04-28 15:35:08Z

13

The matplotlib function 'imshow' gets 3-channel pictures as (h, w, 3) as you can see in the documentation.

It seems that you passed a "batch" of single image (the first dimention) of three channels (second dimention) of the image (h and w are the third and forth dimention).

You need to reshape or view your image (after converting to cpu, try to use:

image1.squeeze().permute(1,2,0)

The result will be an image of the desired shape (128, 128, 3).

The squeeze() function will remove the first dimention. And the premute() function will transpose the dimenstion where the first will shift to the third position and the two other will shift to the beginning.

Also, have a look here for further talk on the GPU and CPU issues: link

Hope that helps.

edited Apr 28, 2020 at 15:35

answered Apr 28, 2020 at 15:27

A. Maman

1,0001 gold badge10 silver badges24 bronze badges

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

1 Comment

Hekes Pekes Over a year ago

Thank you that helped!

Collectives™ on Stack Overflow

Python matplotlib, invalid shape for image data

1 Answer 1

1 Comment

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Linked

Related