https://github.com/nicklhy/AdversarialAutoEncoder

Adversarial AutoEncoder implemented with MXNet
https://github.com/nicklhy/AdversarialAutoEncoder

Last synced: 2 months ago
JSON representation

Adversarial AutoEncoder implemented with MXNet

• Host: GitHub
• URL: https://github.com/nicklhy/AdversarialAutoEncoder
• Owner: nicklhy
• Created: 2016-10-11T02:56:32.000Z (over 8 years ago)
• Default Branch: master
• Last Pushed: 2018-05-18T01:26:12.000Z (over 6 years ago)
• Last Synced: 2024-08-01T22:41:39.888Z (5 months ago)
• Language: Jupyter Notebook
• Size: 797 KB
• Stars: 25
• Watchers: 3
• Forks: 6
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• Awesome-MXNet - AdversarialAutoEncoder

README

        
## Adversarial AutoEncoder

----------------------------

[Adversarial Autoencoder [arXiv:1511.05644]](http://arxiv.org/abs/1511.05644) implemented with [MXNet](https://github.com/dmlc/mxnet).

### Requirements

* MXNet

* numpy

* matplotlib

* scikit-learn

* OpenCV

### Unsupervised Adversarial Autoencoder

Please run aae\_unsupervised.py for model training. Set task to `unsupervised` in visualize.ipynb to display the results. Notice the desired prior distribution of the 2-d latent variable can be one of {gaussian, gaussian mixture, swiss roll or uniform}. In this case, no label info is being used during the training process.

Some results:

p(z) and q(z) with z_prior set to gaussian distribution.

![p(z) gaussian](http://closure11.com/images/post/2016/10/gaussian_unsupervised_pz.png)

![q(z) gaussian](http://closure11.com/images/post/2016/10/gaussian_unsupervised_qz.png)

p(z) and q(z) with z_prior set to 10 gaussian mixture distribution.

![p(z) gaussian](http://closure11.com/images/post/2016/10/gaussian_mixture_unsupervised_pz.png)

![q(z) gaussian](http://closure11.com/images/post/2016/10/gaussian_mixture_unsupervised_qz.png)

p(z) and q(z) with z_prior set to swiss roll distribution.

![p(z) gaussian](http://closure11.com/images/post/2016/10/swiss_roll_unsupervised_pz.png)

![q(z) gaussian](http://closure11.com/images/post/2016/10/swiss_roll_unsupervised_qz.png)

### Supervised Adversarial Autoencoder

Please run aae\_supervised.py for model training. Set task to `supervised` in visualize.ipynb to display the results. Notice the desired prior distribution of the 2-d latent variable can be one of {gaussian mixture, swiss roll or uniform}. In this case, label info of both real and fake data is being used during the training process.

Some results:

p(z), q(z) and output images from fake data with z_prior set to 10 gaussian mixture distribution.

![p(z) gaussian](http://closure11.com/images/post/2016/10/gaussian_mixture_supervised_pz.png)

![q(z) gaussian](http://closure11.com/images/post/2016/10/gaussian_mixture_supervised_qz.png)

![output images from gaussian fake data](http://closure11.com/images/post/2016/10/gaussian_mixture_supervised_output.png)

p(z) and q(z) with z_prior set to swiss roll distribution.

![p(z) gaussian](http://closure11.com/images/post/2016/10/swiss_roll_supervised_pz.png)

![q(z) gaussian](http://closure11.com/images/post/2016/10/swiss_roll_supervised_qz.png)

p(z) and q(z) with z_prior set to 10 uniform distribution.

![p(z) gaussian](http://closure11.com/images/post/2016/10/uniform_supervised_pz.png)

![q(z) gaussian](http://closure11.com/images/post/2016/10/uniform_supervised_qz.png)

### Semi-Supervised Adversarial Autoencoder

Not implemented yet.