https://github.com/yxlao/reverse-gan.pytorch

Precise Recovery of Latent Vectors from Generative Adversarial Networks in PyTorch
https://github.com/yxlao/reverse-gan.pytorch

Last synced: 21 days ago
JSON representation

Precise Recovery of Latent Vectors from Generative Adversarial Networks in PyTorch

• Host: GitHub
• URL: https://github.com/yxlao/reverse-gan.pytorch
• Owner: yxlao
• License: bsd-3-clause
• Created: 2017-05-14T08:14:37.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2017-06-21T04:32:13.000Z (over 7 years ago)
• Last Synced: 2024-10-03T12:42:08.298Z (about 1 month ago)
• Language: Python
• Homepage:
• Size: 12.4 MB
• Stars: 32
• Watchers: 5
• Forks: 2
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# PyTorch-ReverseGAN

PyTorch implementation of

"Precise Recovery of Latent Vectors from Generative Adversarial Networks"

[https://arxiv.org/abs/1702.04782](https://arxiv.org/abs/1702.04782).

Given the generated image `G(z)` with `z` unknown, the goal of ReverseGAN is to

find `z_approx` that approximates `z`. To achieve this, one can find `z_approx`

that minimizes `MSE(G(z_approx), G(z))` through gradient descent.

## Setups

### Dataset

Download the "Align&Cropped Images" from

[http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html](http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html)

and unzip.

### Training

```bash

python dcgan.py --dataset=folder --dataroot=/path/to/dataset --cuda

```

By default, the generated images and saved models will be saved in `dcgan_out`

directory.

### Reverse

After training, run

```bash

python dcgan_reverse.py --clip=stochastic --netG=pre_trained/netG_epoch_10.pth --cuda

```

where `--netG` points to the model saved during training.

### Example results

The following example uses the pre-trained model `pre_trained/netG_epoch_10.pth`

on CelebA aligned dataset.

`G(z)`: the generated image with `z`

![result](pre_trained/g_z.png)

`G(z_approx)`: the generated image with the estimated `z_approx`

![result](pre_trained/g_z_approx.png)

## References

- DCGAN's implementation from

[PyTorch examples](https://github.com/pytorch/examples/tree/master/dcgan).

- Author's [TF implementation](https://github.com/SubarnaTripathi/ReverseGAN).