README.md

GAN-study

Hi, this repository was created as a student of the master's degree in (computer engineering? i'm not sure that is the correct translation :P), and has the propose to help me understend the topic "generative adversial network". Also the english isn´t my strong point, so i hope i can improve as well.

GAN models

Here i present several GAN models and basic classifiers (just to compare results) in format of notebook implemented with tensorflow using the layers API

Current implementation's

1. GAN (original 2014)
2. Conditional GAN
3. NN for MNIST Classification (simple nn, no dropout or other fancies tecniques, just to have baseline score)
4. Auxiliar classifier GAN
5. CNN for MNIST Classification (just to have baseline score)
6. Deep Convolution GAN
7. Condition Deep Convolution GAN

Current/future work

• Condition Deep Convolution GAN
• Semi Supervised GAN (Improved Techniques for Training GANs)
• Information Retrieval GAN

Material used for current study

• GAN - https://arxiv.org/abs/1406.2661
• Conditional GAN - https://arxiv.org/abs/1411.1784
• Auxiliar classifier GAN - https://arxiv.org/abs/1610.09585
• Deep Convolution GAN - https://arxiv.org/abs/1511.06434
• SSGAN - https://arxiv.org/abs/1606.03498
• Book - Hands-On Machine Learning with Scikit-Learn and TensorFlow (first steps)
• Book code - https://github.com/ageron/handson-ml
• Models code examples - https://github.com/wiseodd/generative-models

Material for future study

• ICGAN - https://arxiv.org/abs/1611.06355
• EBGAN - https://arxiv.org/abs/1609.03126
• BEGAN - https://blog.heuritech.com/2017/04/11/began-state-of-the-art-generation-of-faces-with-generative-adversarial-networks/
• BEGAN - (paper)
• IRGAN - (paper)

Requirements

• jupyter notebook
• tensorflow (1.4, 1.5 (soon)) (All the code was run on GPU version (but CPU should work to))
• numpy
• matplotlib