https://stackoverflow.com/questions/42702477/how-to-make-my-own-mnist-dataset

how to make my own mnist dataset?

Tensorflow mnist download mnist dataset for English digits , I'm working with mnist for Arabic digits and I have JPG !

how to convert JPG to my own " t10k-images-idx3-ubyte.gz " and " t10k-lables-idx1-ubyte.gz"?

help me please , thanks.

shareimprove this question
 
    
What are you trying to do? – Mad Physicist Mar 9 at 18:16
    
I'm working on arabic handwritten characters recognition using deep learning –  adeeb Mar 9 at 20:53

You don't need to convert them into byte and tar format. You can just feed them into the placeholder.

img = tf.placeholder(tf.float32, shape=(None))

...

digit = scipy.ndimage.imread("arabic_digit.jpg")
feed_dict = {img:digit}
sess.run(model, feed_dict=feed_dict)

if you have multiple digits just load them into a numpy array before feeding it into the feed dict.

digits = []
for digit_location in digits_list:
    digits.append(scipy.ndimage.imread(digit_location))
digits = np.asarray(digits)
feed_dict = {img:digits}
sess.run(model, feed_dict=feed_dict)
shareimprove this answer
 
    
thanks steven for quick response <3 –  adeeb Mar 9 at 20:50
1  
If there's nothing else that you need please mark the question as solved. Thanks. – Steven Mar 9 at 21:47
    
what is digits_list? –  adeeb Mar 10 at 11:09
    
steven, thanks for comments, but I have training dataset and testing dataset, how to implement it? –  adeebMar 10 at 12:57
    
How to implement what? I would need you to be specific. digits_list is a list of locations of the images you wish to load. If you mean a model like mnist you can see here tensorflow.org/get_started/mnist/pros and heregithub.com/tensorflow/tensorflow/blob/master/tensorflow/… – Steven Mar 10 at 16:31