https://github.com/tatsy/normalizing-flows-pytorch

PyTorch implementations of normalizing flow and its variants.
https://github.com/tatsy/normalizing-flows-pytorch

Last synced: 3 months ago
JSON representation

PyTorch implementations of normalizing flow and its variants.

• Host: GitHub
• URL: https://github.com/tatsy/normalizing-flows-pytorch
• Owner: tatsy
• Created: 2020-12-14T10:47:59.000Z (almost 4 years ago)
• Default Branch: master
• Last Pushed: 2021-05-14T07:58:11.000Z (over 3 years ago)
• Last Synced: 2024-07-04T00:58:59.749Z (4 months ago)
• Language: Python
• Size: 7.36 MB
• Stars: 70
• Watchers: 5
• Forks: 8
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
Normalizing Flows by PyTorch

===

[![Codacy Badge](https://app.codacy.com/project/badge/Grade/95bd163ba34a495789e073b94bd2da3d)](https://www.codacy.com/gh/tatsy/normalizing-flows-pytorch/dashboard?utm_source=github.com&utm_medium=referral&utm_content=tatsy/normalizing-flows-pytorch&utm_campaign=Badge_Grade)

> PyTorch implementations of the networks for normalizing flows.

Models

---

Currently, following networks are implemented.

* Planar flow

  * Rezende and Mohamed 2015, "Variational Inference with Normalizing Flows," [[arXiv]](https://arxiv.org/abs/1505.05770)

* RealNVP

  * Dinh et al., 2016, "Density Estimation using Real NVP," [[arXiv]](https://arxiv.org/abs/1605.08803)

* Glow

  * Kingma and Dhariwal 2018, "Glow: Generative Flow with Invertible 1x1 Convolutions," [[arXiv]](https://arxiv.org/abs/1807.03039v2) [[code]](https://github.com/openai/glow)

* Flow++

  * Ho et al., 2019, "Flow++: Improving Flow-Based Generative Models with Variational Dequantization and Architecture Design," [[arXiv]](https://arxiv.org/abs/1902.00275) [[code]](https://github.com/aravindsrinivas/flowpp)

* MAF

  * Papamakarios et al., 2017, “Masked Autoregressive Flow for Density Estimation,” [[arXiv]](https://arxiv.org/abs/1705.07057)

* Residual Flow

  * Behrmann et al., 2018, "Residual Flows for Invertible Generative Modeling," [[arXiv]](https://arxiv.org/abs/1906.02735) [[code]](https://github.com/rtqichen/residual-flows)

* FFJORD

  * Grathwohl et al., 2018, "FFJORD: Free-form Continuous Dynamics for Scalable Reversible Generative Models," [[arXiv]](https://arxiv.org/abs/1810.01367) [[code]](https://github.com/rtqichen/ffjord)

**Note:**

This repository is for easier understanding of the above networks. Therefore, you should use official source cods if provided.

Setup

---

### Anaconda

By Anaconda, you can easily setup the environment using `environment.yml`.

```shell

$ conda env create -f environment.yml

```

### Pip

If you use `pip` or other tools, see the dependencies in [`environment.yml`](./environment.yml)

Run

---

This repo uses `hydra` to manage hyper parameters in training and evaluation. See [`configs`](./configs) folder to check the parameters for each network.

```shell

$ python main.py \

    network=[planar, realnvp, glow, flow++, maf, resflow, ffjord]\

    run.distrib=[circles, moons, normals, swiss, s_curve, mnist, cifar10]

```

**Note:**

Currently, I tested the networks only for 2D density transformation. So, results for 3D densities (`swiss` and `s_curve`) and images (`mnist` and `cifar10`) could be what you expect.

Results

---

See [`results/README.md`](./results/README.md) for more results.

### Real NVP

| Target | Reproduced | Training |

|:--------:|:------------:|:----:|

| ![](./results/density/realnvp/y_data_020000.jpg) | ![](./results/density/realnvp/y_sample_020000.jpg) | ![](./results/density/realnvp/movie.gif) |

Copyright

---

MIT License (c) 2020, Tatsuya Yatagawa