CODEWITHSUNDEEP
imagetensorflow
pangdan
pangdan

Reputation: 61

Confused image after tf.image.convert_image_dtype(image, dtype=tf.uint8 )

I have read a grayscale image and looks normal, but the image looks very strange after converting data type by tf.image.convert_image_dtype(). I don't know what happened, appreciate any help.

... ...
uint_inputs = tf.image.convert_image_dtype(inputs, dtype=tf.uint8, saturate=False)
... ...
with sv.managed_session() as sess:
    inputs, uint_inputs = sess.run([inputs, uint_inputs])
    f ,axis = plt.subplots(1, 2, figsize=(4, 2))
    axis[0].imshow(np.squeeze(inputs), cmap='gray')
    axis[1].imshow(np.squeeze(uint_inputs), cmap='gray')
    plt.show()

Results after running as follows, enter image description here

Upvotes: 1

Views: 443

Answers (1)

pangdan
pangdan

Reputation: 61

After a long struggle, I found the inputs must be in [0, 1) if they are float type in tf.image.convert_image_dtype(inputs, dtype=tf.uint8, saturate=False). Thus, I just scale the inputs by inputs = inputs/255 before they are feed into the tf.image.convert_image_dtype pipline.

Upvotes: 1

Related Questions