Using the svhn classifier to predict a single image

[bambrozio]

Hi! Thank you very much for sharing this work. I'm learning a lot from it! (I'm new on ML/DL)
I'm trying to use your model to predict single images, but no success. Would you have some minutes to help to implement an interface for that?
I'm trying here, but I'm facing:
Cannot feed value of shape (32, 32, 4) for Tensor 'input/x:0', which has shape '(?, 32, 32, 1)'.
Would be nice if you could share a python script where we can input an image, and it does crop and tries to predict it...
Thanks!