https://github.com/alleninstitute/u-dagan
Unsupervised data augmentation using GANs.
https://github.com/alleninstitute/u-dagan
Last synced: 11 months ago
JSON representation
Unsupervised data augmentation using GANs.
- Host: GitHub
- URL: https://github.com/alleninstitute/u-dagan
- Owner: AllenInstitute
- Created: 2020-07-13T18:32:42.000Z (almost 6 years ago)
- Default Branch: master
- Last Pushed: 2020-09-16T05:45:56.000Z (almost 6 years ago)
- Last Synced: 2025-04-11T20:11:25.937Z (about 1 year ago)
- Language: Python
- Size: 11.6 MB
- Stars: 7
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Unsupervised Adversarial Augmenter/Generator Networks
This repository contains an implementation of an unspervised adverserial framework for both data augmentation and generarion.
## Table of contents
* [Generator](#generator)
* [Augmenter](#augmenter)
* [Usage](#usage)
## Generator
combining a variational autoencoder (VAE) with a generative adversarial network (GAN), we introduce a VAE-GAN netowrk that leverages unsupervised representation learning and data sample reconstruction.
The network architecture for the developed VAE-GAN.
## Augmenter
Given the adversarial training proposed for GANs, here we introduce an augmentation network thatgenerates multiple nonidentical augmented samples with identical class labels, called U-DAGAN.
The schematic of the proposed architecture for unsupervised data augmentation and the augmenter's architecture.
### Example
#### MNIST
#### snRNA-seq (FACS)
## Usage
Each **generator** and **augmenter** folder contains code for training the network and generating fakes samples.
To train the model you can use `run.py` to train the network and `*augmenter.py`/`*augmenter.py` to generate fake samples.