CODEWITHSUNDEEP
keraskeras-layer
Oblomov
Oblomov

Reputation: 9645

Keras Cropping2D changes color channel

I tried to visualize the effect of keras cropping 2D using the following code snippet:

from keras import backend as K
from keras.layers.convolutional import Cropping2D
from keras.models import Sequential
# with a Sequential model
model = Sequential()
model.add(Cropping2D(cropping=((22, 0), (0, 0)), input_shape=(160, 320, 3)))
cropping_output = K.function([model.layers[0].input],
                                  [model.layers[0].output])
cropped_image = cropping_output([image[None,...]])[0]
compare_images(image,
               cropped_image.reshape(cropped_image.shape[1:]))

Here is the plotting function:

def compare_images(left_image, right_image):    
    print(image.shape)
    f, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
    f.tight_layout()
    ax1.imshow(left_image)
    ax1.set_title('Shape '+ str(left_image.shape),
                  fontsize=50)
    ax2.imshow(right_image)
    ax2.set_title('Shape '+ str(right_image.shape)
                  , fontsize=50)
    plt.show()

The result is

enter image description here

Obviously, the color channel has been changed. But why? Is there an error in my code or could that be a keras bug?

Upvotes: 1

Views: 1640

Answers (1)

indraforyou
indraforyou

Reputation: 9099

Its not a Keras bug. Tensors are usually of float32 type so when the output is evaluated they are also of float32 type. You need to convert the image data to uint8 type before displaying.

ax2.imshow(np.uint8(right_image))

in compare_images should display the image correctly.

Upvotes: 1

Related Questions