using ImageDataGenerator, matplotlib throws TypeError: Invalid shape (1, 256, 256, 3) for image data-CodePudding

I have 15 images of cars, and using data augmentation, I want to create a dataset out of them. However, when I use ImageDataGenerator from Keras and try to plot the generated images, I'm getting an error that says

TypeError: Invalid shape (1, 256, 256, 3) for image data.

I am attaching the code as well, please let me know how I can fix this.

datagen = ImageDataGenerator(rescale=1./255, zoom_range=0.1, rotation_range=25, width_shift_range=0.1, height_shift_range=0.1, shear_range=0.1, horizontal_flip=True)

ite = datagen.flow_from_directory("Car Images", batch_size=1)

for i in range(9):

    # define subplot
    plt.subplot(330   1   i)

    # generate batch of images
    batch = ite.next()

    # convert to unsigned integers for viewing
    image = batch[0].astype('uint8')

    # plot raw pixel data
    plt.imshow(image)

# show the figure
plt.show()

The error points to the plt.imshow() line.

CodePudding user response：

You need to reshape the image, try this.

datagen = ImageDataGenerator(rescale=1./255, zoom_range=0.1, rotation_range=25, width_shift_range=0.1, height_shift_range=0.1, shear_range=0.1, horizontal_flip=True)

ite = datagen.flow_from_directory("Car Images", batch_size=1)

for i in range(9):

    # define subplot
    plt.subplot(330   1   i)

    # generate batch of images
    batch = ite.next()

    # convert to unsigned integers for viewing
    image = batch[0].astype('uint8')

    image = np.reshape(256,256,3)

    # plot raw pixel data
    plt.imshow(image)

# show the figure
plt.show()

CodePudding user response：

For Invalid Shape error, you should remove batch dimension. So, reshaping or using np.squeeze() is required.

Furthermore, since you are rescaling your images by 1./255, images data are in range [0,1], and converting them to uint8 makes all of them zero. So change your last 2 lines in for loop like this:

image = batch[0]                  #remove astype('uint8')
# plot raw pixel data
plt.imshow(np.squeeze(image))     #remove batch dimension