davidtellez / dcgan_mnist64

Implementation of a Deep Convolutional Generative Adversarial Network to generate realistic MNIST digits at 64x64 resolution in Tensorflow.

DCGAN MNIST 64x64

Implementation of Deep Convolutional Generative Adversarial Networks to generate realistic MNIST digits at 64x64 resolution in Tensorflow. After 40 epochs of training, you should get these results:

Usage

1. Launch jupyter notebook and browse to http://localhost:8888/
2. Open [DCGAN MNIST tutorial.ipynb](DCGAN MNIST tutorial.ipynb)
3. Run all to start training
4. Use Tensorboard to monitor the training: tensorboard --logdir YOUR_DIR and browse to http://localhost:6006/

Requisites

• Anaconda Python 3.5
• Tensorflow 0.12.0-rc1
• GPU for fast training (although CPU-only is supported)

References

• Generative Adversarial Networks
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• Image Completion with Deep Learning in TensorFlow