Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

https://github.com/tensorflow/gan

Tooling for GANs in TensorFlow
https://github.com/tensorflow/gan

Last synced: about 1 month ago
JSON representation

Tooling for GANs in TensorFlow

Lists

README 

        
# TensorFlow-GAN (TF-GAN)



TF-GAN is a lightweight library for training and evaluating

[Generative Adversarial Networks (GANs)](https://arxiv.org/abs/1406.2661).



*   Can be installed with `pip` using `pip install tensorflow-gan`, and used

    with `import tensorflow_gan as tfgan`

*   [Well-tested examples](https://github.com/tensorflow/gan/tree/master/tensorflow_gan/examples/)

*   [Interactive introduction to TF-GAN](https://github.com/tensorflow/gan/blob/master/tensorflow_gan/examples/colab_notebooks/tfgan_tutorial.ipynb) in



## Structure of the TF-GAN Library



TF-GAN is composed of several parts, which are designed to exist independently:



*   [Core](https://github.com/tensorflow/gan/tree/master/tensorflow_gan/python/train.py):

    the main infrastructure needed to train a GAN. Set up training with any

    combination of TF-GAN library calls, custom-code, native TF code, and other

    frameworks

*   [Features](https://github.com/tensorflow/gan/tree/master/tensorflow_gan/python/features/):

    common GAN operations and normalization techniques, such as instance

    normalization and conditioning.

*   [Losses](https://github.com/tensorflow/gan/tree/master/tensorflow_gan/python/losses/):

    losses and penalties, such as the Wasserstein loss, gradient penalty, mutual

    information penalty, etc.

*   [Evaluation](https://github.com/tensorflow/gan/tree/master/tensorflow_gan/python/eval/):

    standard GAN evaluation metrics. Use `Inception Score`, `Frechet Distance`,

    or `Kernel Distance` with a pretrained Inception network to evaluate your

    unconditional generative model. You can also use your own pretrained

    classifier for more specific performance numbers, or use other methods for

    evaluating conditional generative models.

*   [Examples](https://github.com/tensorflow/gan/tree/master/tensorflow_gan/):

    simple examples on how to use TF-GAN, and more complicated state-of-the-art

    examples



## Who uses TF-GAN?



Numerous projects inside Google. The following are some published papers that

use TF-GAN:



*   [Self-Attention Generative Adversarial Networks](https://arxiv.org/abs/1805.08318)

*   [Large Scale GAN Training for High Fidelity Natural Image Synthesis](https://arxiv.org/abs/1809.11096)

*   [GANSynth: Adversarial Neural Audio Synthesis](https://arxiv.org/abs/1902.08710)

*   [Boundless: Generative Adversarial Networks for Image Extension](http://arxiv.org/abs/1908.07007)

*   [NetGAN: Generating Graphs via Random Walks](https://arxiv.org/abs/1803.00816)

*   [Discriminator rejection sampling](https://arxiv.org/abs/1810.06758)

*   [Generative Models for Effective ML on Private, Decentralized Datasets](https://arxiv.org/pdf/1911.06679.pdf)

*   [Semantic Pyramid for Image Generation](https://arxiv.org/abs/2003.06221)

*   [GAN-Mediated Cell Images Batch Equalization](https://www.biorxiv.org/content/10.1101/2020.02.07.939215v1.full)



The framework [Compare GAN](https://github.com/google/compare_gan) uses TF-GAN,

especially the evaluation metrics.

[Their papers](https://github.com/google/compare_gan#compare-gan) use TF-GAN to

ensure consistent and comparable evaluation metrics. Some of those papers are:



*   [Are GANs Created Equal? A Large-Scale Study](https://arxiv.org/abs/1711.10337)

*   [The GAN Landscape: Losses, Architectures, Regularization, and Normalization](https://arxiv.org/abs/1807.04720)

*   [Assessing Generative Models via Precision and Recall](https://arxiv.org/abs/1806.00035)

*   [High-Fidelity Image Generation With Fewer Labels](https://arxiv.org/abs/1903.02271)



## Training a GAN model



Training in TF-GAN typically consists of the following steps:



1.  Specify the input to your networks.

1.  Set up your generator and discriminator using a `GANModel`.

1.  Specify your loss using a `GANLoss`.

1.  Create your train ops using a `GANTrainOps`.

1.  Run your train ops.



At each stage, you can either use TF-GAN's convenience functions, or you can

perform the step manually for fine-grained control.



There are various types of GAN setup. For instance, you can train a generator to

sample unconditionally from a learned distribution, or you can condition on

extra information such as a class label. TF-GAN is compatible with many setups,

and we demonstrate in the well-tested

[examples directory](https://github.com/tensorflow/gan/tree/master/tensorflow_gan/examples/)



## Maintainers



*   (Documentation) David Westbrook, [email protected]

*   Joel Shor, [email protected], [github](https://github.com/joel-shor)

*   Aaron Sarna, [email protected], [github](https://github.com/aaronsarna)

*   Yoel Drori, [email protected], [github](https://github.com/yoeldr)



## Authors



*   Joel Shor, [email protected], [github](https://github.com/joel-shor)