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.