These are the Modules for Invertible Neural Networks (MoINN)

It is mainly a Tensorflow 2 implementation of the FrEIA framework: https://github.com/VLL-HD/FrEIA

Currently implemented coupling blocks are:

• NICE [1]
• Real NVP [2]
• GLOW [3]
• GIN [4]
• SPADE [5]
• All in one Block (FrEIA)

Installation

Dependencies

Package	Version
Python	3.6 - 3.8
Tensorflow	>= 2.1.0
Numpy	>= 1.15.0
Scipy	>= 1.5

Download + Install

To just install the modules for usage:

pip install git+https://github.com/ramonpeter/MoINN.git

For development:

# clone the repository
git clone https://github.com/ramonpeter/MoINN.git
# then install in dev mode
cd MoINN
pip install --editable .

References

[1]	Dinh et al., "NICE: Non-linear Independent Components Estimation", 1410.8516 [cs.LG]

[2]	Dinh et al., “Density estimation using Real NVP,” 1605.08803 [cs.LG]

[3]	Kingma and Dhariwal, “Glow: Generative Flow with Invertible 1x1 Convolutions”, 1807.03039 [cs.LG]

[4]	Sorrenson et al., "Disentanglement by Nonlinear ICA with General Incompressible-flow Networks (GIN)", 2001.04872 [cs.LG]

[5]	Park et al., "Semantic Image Synthesis with Spatially-Adaptive Normalization", 1903.07291 [cs.LG]