TensorFlow - Show image from MNIST DataSet

Question

I'm trying to learn TensorFlow and I implemented the MNIST example from the the following link: http://openmachin.es/blog/tensorflow-mnist I want to be able to actually view the training/test images. So I'm trying to add code that will show the first train picture of the first batch:

x_i = batch_xs[0]
image = tf.reshape(x_i,[28,28])

Now, because the Data is in float32 type (with values in [0,1] range), I tried to convert it to uint16 and then to encode it to png in order to show the image. I tried using tf.image.convert_image_dtype and tf.image.encode_png, but with no success. Can you guys please help me understand how can I convert the raw Data to an image and show the image?

Lonilonier

Possible duplicate of Display MNIST image using matplotlib – Birl

Answer 1

After reading the tutorial you can do it all in numpy no need for TF:

import matplotlib.pyplot as plt
first_array=batch_xs[0]
#Not sure you even have to do that if you just want to visualize it
#first_array=255*first_array
#first_array=first_array.astype("uint8")
plt.imshow(first_array)
#Actually displaying the plot if you are not in interactive mode
plt.show()
#Saving plot
plt.savefig("fig.png")

You can also use PIL or whatever visualization tool you are into.

Ecclesiasticus

I want to save the image via TensorFlow in some windows readable format, and then open the image in windows image viewer. How can I do that? – Lonilonier

You just have to use plt.savefig to save the plot after the lines I wrote, but your original question was only for display. – Ecclesiasticus

Maybe you are unfamiliar with matplotlib I will modify my answer to make it more suitable to your needs. – Ecclesiasticus

I dont have matplotlib available, this is why I wanted to convert the image (using TensorFlow) into a format which I can open later in windows. Is this possible? – Lonilonier

#903261 – Ecclesiasticus

Not having matplotlib/numpy installed when using Python for scientific purposes to me is a bit like playing Tennis without a racket of course you could manage to bounce the ball back once or twice but you will soon get tired of this. If you still persist you can always use tf.image_summary jointly with Tensorboard for visualization only. – Ecclesiasticus

The link I posted provides several way not to use matplotlib (by using PIL, or even pure Python even if it looks awfull). – Ecclesiasticus

Answer 2

X = X.reshape([28, 28]);
plt.gray()
plt.imshow(X)

this works.

Turtleneck

Answer 3

On top of the codes in the tutorial MNIST for ML beginners, you can visualize the image in the mnist dataset:

import matplotlib.pyplot as plt
batch = mnist.train.next_batch(1)
plotData = batch[0]
plotData = plotData.reshape(28, 28)
plt.gray() # use this line if you don't want to see it in color
plt.imshow(plotData)
plt.show()

enter image description here

Svetlanasvoboda

Answer 4

Pass a numpy array representing an MNIST image to the function below and it will display a figure using matplotlib.

def displayMNIST(imageAsArray):
    imageAsArray = imageAsArray.reshape(28, 28);
    plt.imshow(imageAsArray, cmap='gray')
    plt.show()

Margoriemargot

Unless you explain your answer and how it's a better version of the function, it will likely get down-voted or deleted. You have to provide a little context. Welcome to Stack Overflow. Thanks for answering questions. – Milligram

While this might answer the authors question, it lacks some explaining words and/or links to documentation. Raw code snippets are not very helpful without some phrases around them. You may also find how to write a good answer very helpful. Please edit your answer - From Review – Expectant

Answer 5

In tensorflow 2.0:

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

(x_train, y_train), (x_test, y_test) = mnist.load_data()

plt.imshow(x_train[0], cmap='gray_r')

Holladay

Answer 6

We can use subplots of matplotlib.pyplot

 # plotting the first 9 images in the train set of MNIST
 fig , axs = plt.subplots(3, 3)
 cnt = 0
 for i in range(3):
     for j in range(3):
         axs[i, j].imshow(X_train[cnt])
         cnt += 1

Perry