尝试显示从 rgb 转换的灰度图像时出错

Question

I'm doing a project in deeplearning where I'll train a classifier using the SVHN dataset.

Initially, I had train_images and test_images with shapes (73257, 32, 32, 3) and (26032, 32, 32, 3), respectively and they are normally displayed using plt.imshow().

I converted the images to greyscale using the following code:

f = lambda img: tf.expand_dims(np.average(img, axis=-1), axis=-1).numpy()

train_images = f(train_images)
test_images = f(test_images)

I tried, as well:

train_images = tf.image.rgb_to_grayscale(train_images).numpy()
test_images = tf.image.rgb_to_grayscale(test_images).numpy()

On both cases, the output gives images with shape (32, 32, 1) as items in each of the train_images, test_images - the new shapes are, therefore, (73257, 32, 32, 3) and (26032, 32, 32, 1), respectively.

But, for some reason, when I try to display these new greyscale images, I get the following error : TypeError: Invalid dimensions for image data.

I have no idea why.

Answer 1

I solved the problem by using the following function for transforming to grayscale:

f = lambda img: np.mean(img, axis=-1, keepdims=True)

Then displayed the image using:

img = np.squeeze(array)
plt.imshow(img, cmap='Greys')

Where array is any item on train_images or test_images.