Tensorflow implementation of conditional Generative Adversarial Networks (cGAN) [1] and conditional Deep Convolutional Generative Adversarial Networks (cDCGAN) for MANIST [2] dataset.

• you can download
  • MNIST dataset: http://yann.lecun.com/exdb/mnist/

• cGAN

• Generate using fixed noise (fixed_z_)

cGAN	cDCGAN

• MNIST vs Generated images

MNIST	cGAN after 100 epochs	cDCGAN after 30 epochs

• Training loss

cGAN	cDCGAN

• Learning time
  • MNIST cGAN - Avg. per epoch: 3.21 sec; Total 100 epochs: 1800.37 sec
  • MNIST cDCGAN - Avg. per epoch: 53.07 sec; Total 30 epochs: 2072.29 sec

• Windows 7
• GTX1080 ti
• cuda 8.0
• Python 3.5.3
• tensorflow-gpu 1.2.1
• numpy 1.13.1
• matplotlib 2.0.2
• imageio 2.2.0

[1] Mirza, Mehdi, and Simon Osindero. "Conditional generative adversarial nets." arXiv preprint arXiv:1411.1784 (2014).

(Full paper: https://arxiv.org/pdf/1411.1784.pdf)

[2] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. "Gradient-based learning applied to document recognition." Proceedings of the IEEE, 86(11):2278-2324, November 1998.