https://github.com/aravindsrinivas/flowpp

Code for reproducing Flow ++ experiments
https://github.com/aravindsrinivas/flowpp

deep-learning density-estimation generative-models neural-networks normalizing-flows tensorflow

Last synced: 3 months ago
JSON representation

Code for reproducing Flow ++ experiments

• Host: GitHub
• URL: https://github.com/aravindsrinivas/flowpp
• Owner: aravindsrinivas
• License: mit
• Created: 2018-10-02T01:21:36.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2019-04-27T21:40:11.000Z (over 5 years ago)
• Last Synced: 2024-04-30T07:32:40.888Z (7 months ago)
• Topics: deep-learning, density-estimation, generative-models, neural-networks, normalizing-flows, tensorflow
• Language: Python
• Homepage:
• Size: 103 KB
• Stars: 183
• Watchers: 6
• Forks: 27
• Open Issues: 4
• Metadata Files:
  • Readme: README.md
  • License: LICENSE.md

Awesome Lists containing this project

• awesome-face-related-list - Flow++

README

        
# Flow++: Improving Flow-Based Generative Models with Variational Dequantization and Architecture Design 

This repository contains Tensorflow implementation of experiments from the paper [Flow++: Improving Flow-Based Generative Models with Variational Dequantization and Architecture Design](https://arxiv.org/abs/1902.00275) - [Jonathan Ho](http://www.jonathanho.me/), [Xi Chen](http://peterchen.io/), [Aravind Srinivas](https://people.eecs.berkeley.edu/~aravind/), [Yan Duan](http://rockyduan.com/), [Pieter Abbeel](https://people.eecs.berkeley.edu/~pabbeel/)

# Dependencies

* python3.6 

* Tensorflow v1.10.1 

* [horovod v0.14.1](https://github.com/uber/horovod)

[Horovod GPU setup instructions](https://github.com/uber/horovod/blob/master/docs/gpus.md)

# Usage Instructions

We trained our models using 8 GPUs with data-parallelism using Horovod. 

# CIFAR 10 

```

mpiexec -n 8 python3.6 run_cifar.py

```

# Imagenet 

## Data for Imagenet Experiments: 

Script to create dataset [here](https://github.com/aravind0706/flowpp/blob/master/flows_imagenet/create_imagenet_benchmark_datasets.py)

## Imagenet 32x32

```

mpiexec -n 8 python3.6 -m flows_imagenet.launchers.imagenet32_official

```

# Imagenet 64x64

```

mpiexec -n 6 python3.6 -m flows_imagenet.launchers.imagenet64_official

mpiexec -n 6 python3.6 -m flows_imagenet.launchers.imagenet64_5bit_official

```

# CelebA-HQ 64x64 

## Data: 

Download links in [README](https://github.com/aravind0706/flowpp/tree/master/flows_celeba)

```

mpiexec -n 8 python3.6 -m flows_celeba.launchers.celeba64_5bit_official

mpiexec -n 8 python3.6 -m flows_celeba.launchers.celeba64_3bit_official

```

# Contact

Please open an issue

# Credits

flowpp was originally developed by Jonathan Ho (UC Berkeley), Peter Chen (UC Berkeley / covariant.ai), Aravind Srinivas (UC Berkeley), Yan Duan (covariant.ai), and Pieter Abbeel (UC Berkeley / covariant.ai).