TypeError: Image data cannot be converted to float after tf.image.per_image_standardization(x)

Question

I am getting the following error at plt.imshow

TypeError: Image data cannot be converted to float

For this code:

import keras
import tensorflow as tf
import matplotlib.pyplot as plt
mnist = keras.datasets.mnist

(train_images, train_labels), (test_images, test_labels) = mnist.load_data()

def preprocess(x):
    x = tf.image.per_image_standardization(x)
    return x

train_images = preprocess(train_images)
test_images = preprocess(test_images)

plt.figure()
plt.imshow(train_images[1])
plt.colorbar()
plt.grid(False)
plt.show()

Any ideas why this is happening? Thanks!

Answer 1

In your script, train_images do not contain actual data but are merely placeholder Tensors:

train_images[1]
<tf.Tensor 'strided_slice_2:0' shape=(28, 28) dtype=float32>

The simplest solution would be to enable eager execution at the top of your script:

tf.enable_eager_execution()

This means that at runtime, the tensors will actually contain the data that you are trying to plot:

train_images[1]
<tf.Tensor: id=95, shape=(28, 28), dtype=float32, numpy=
array([[-0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 ,
        -0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 ,
        -0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 ,
        -0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 ,
        -0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 , -0.4250042 ,
        -0.4250042 , -0.4250042 , -0.4250042 ], # etc

Which should solve your error. You can read more about eager execution on TF's website .

Alternatively, you can also make the plot by actually evaluating the image tensors in a session:

with tf.Session() as sess:
    img = sess.run(train_images[1])
    plt.figure()
    plt.imshow(img)
    plt.colorbar()
    plt.grid(False)
    plt.show()