tf-gen-models 0.0.12

Ready to use implementations of state-of-the-art generative models in TensorFlow 2

Ready to use implementations of state-of-the-art generative models in TensorFlow 2

Installation

Dependencies

tf-gen-models requires:

• Python (>= 3.7, < 3.10)
• TensorFlow (>= 2.5)
• Matplotlib (>= 3.4)
• Pillow (>= 8.0)

---

The tf-gen-models package is built upon TensorFlow 2. See the TensorFlow install guide for the pip package while, to enable GPU support, the use Docker container is recommended. Alternatively, GPU-enabled TensorFlow can be easily installed using the tensorflow-gpu package on conda-forge.

User installation

If you already have a working installation of TensorFlow 2 (preferably with the GPU support enabled), the easiest way to install tf-gen-models is using pip:

pip install tf-gen-models

Available generative models

Generative models	Implementation	Notebooks	Trends
GAN	✔️	🛠️
VAE	❌	❌
Norm Flow	❌	❌

Generative Adversarial Networks

Algorithms	Implementation	Conditioning*	Notebooks	Paper
GAN	✔️	🛠️	✔️	arXiv:1406.2661
BceGAN	✔️	❌	✔️
WGAN	✔️	❌	✔️	arXiv:1701.07875
WGAN_GP	✔️	❌	✔️	arXiv:1704.00028
CramerGAN	✔️	❌	✔️	arXiv:1705.10743
WGAN_ALP	✔️	❌	🛠️	arXiv:1907.05681

*Referring to the conditional version of GANs proposed in arXiv:1411.1784.

Variational Autoencoder

Planned for release v0.2.0

Normalizing Flows

Planned for release v0.2.0

Jupyter notebooks

• MNIST generation with GANs [GAN] [BceGAN] [WGAN] [WGAN-GP] [CramerGAN]

License

MIT License