Inspired by the paper on Stack GAN https://arxiv.org/pdf/1612.03242.pdf which generates realistic images from text, I have tried to find relation between image and text and generating both of them simultaneously using Generative Adversarial Networks.
Following is the architecture of my model
Programming language used: python 3.5
Libraries used: tensorflow 1.4(gpu version), numpy, matplotlib and pickle
Below are some snippets of the results of images and text generated during training. Please compare both image and text generated to know the relation between them.
epoch27,Generated Nos.
['8', '3', '3', '8', '9', '8', '9', '0', '3', '6',
'0', '6', '0', '6', '1', '0', '8', '8', '8', '0',
'1', '3', '2', '9', '3', '1', '8', '9', '3', '1',
'8', '6', '3', '1', '8', '1', '3', '8', '1', '8',
'6', '8', '3', '8', '1', '8', '3', '1', '0', '1',
'3', '9', '0', '0', '1', '8', '3', '1', '9', '1',
'3', '9', '8', '8', '8', '8', '6', '8', '9', '1',
'8', '6', '0', '9', '0', '9', '6', '1', '6', '8',
'3', '8', '3', '0', '6', '8', '8', '8', '3', '1',
'9', '3', '6', '8', '6', '8', '0', '3', '8', '8']
epoch31,Generated Nos.
['8', '6', '3', '8', '8', '8', '8', '0', '3', '8',
'8', '6', '8', '6', '2', '0', '4', '3', '8', '0',
'8', '8', '8', '8', '3', '2', '8', '1', '6', '3',
'8', '6', '3', '1', '8', '1', '6', '8', '2', '8',
'8', '8', '8', '8', '3', '0', '3', '2', '8', '0',
'2', '8', '0', '0', '8', '8', '3', '2', '8', '1',
'3', '9', '8', '1', '8', '4', '6', '8', '0', '1',
'8', '8', '0', '2', '0', '8', '4', '3', '8', '8',
'6', '8', '8', '0', '8', '3', '8', '8', '8', '2',
'8', '3', '6', '8', '4', '4', '0', '6', '8', '8']