Python >= 3.6
Tensorflow
PILLOW == 8.1.0
Numpy == 1.19.5

Input_layer_size = 784 or (1, 28, 28) for image of dimension 28 x 28
Hidden_layer1_size = 512  
Hidden_layer2_size = 256  
Hidden_layer3_size = 128  
Output_layer_size = 10 for 0 - 9 digits

You can use the preprocessed model created from the script tensorflow v2 with an accuracy of 97 percent to 
test your own sample images which is stored in the directory preprocessed_model, you can store your test 
images in the Stock folder and use it in both the scripts.

You don't need to download any dataset and load it extensively just run either of the script and it will 
automatically download and/or load the dataset and you're good to go.