https://github.com/raphaelsenn/gan

PyTorch reimplementation of "Generative Adversarial Networks" (Goodfellow et al., 2014).
https://github.com/raphaelsenn/gan

Last synced: 8 months ago
JSON representation

PyTorch reimplementation of "Generative Adversarial Networks" (Goodfellow et al., 2014).

• Host: GitHub
• URL: https://github.com/raphaelsenn/gan
• Owner: raphaelsenn
• License: mit
• Created: 2025-09-08T11:45:34.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2025-09-17T08:02:53.000Z (9 months ago)
• Last Synced: 2025-09-17T10:12:32.811Z (9 months ago)
• Language: Python
• Size: 2.06 MB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Generative Adversarial Networks (Goodfellow et al., 2014)

This code implements the fully connected generator and discriminator for the MNIST dataset as described in Goodfellow et al. (2014), ["Generative Adversarial Networks"](https://arxiv.org/abs/1406.2661).

I extend the experiment by applying the architecture to the Fashion-MNIST dataset.

![gan_objective](./assets/gan_objective.png)

*Figure: The original GAN objective, where G denotes the generator and D the discriminator (Goodfellow et al., 2014).*

Original code and hyperparameters from the paper can be found [here](https://github.com/goodfeli/adversarial).

## MNIST (fully connected model)

#### Model Architecture:

Generator:

* Input layer: 100 units (Uniform[-1, 1] noise)

* Dense layer: 1200 units, ReLU

* Dense layer: 1200 units, ReLU

* Output layer: 784 units, Sigmoid

* Learnable parameters: 2503984

Discriminator:

* Input layer: 784 units (flattend 28x28 gray-scale image)

* Maxout layer: 240 units, 5 pieces

* Maxout layer: 240 units, 5 pieces

* Output layer: 1 unit, Linear

* Learnable parameters: 1231441

#### Hyperparameters/Settings:

* Epochs: 100  

* Batch size: 128

* Objective: Binary Cross-Entropy Loss with Logits

* Generator optimizer: Adam

* Discriminator optimizer: Adam

* Generator learning rate: 0.002  

* Discriminator learning rate: 0.002  

* Generator betas: (0.5, 0.999)  

* Discriminator betas: (0.5, 0.999)  

#### Training Data:

![MNIST_Training_Data](./assets/mnist_train.png)

#### GAN Training Losses:

![MNIST_Loss](./assets/mnist_loss.png)

#### Fake Data:

![MNIST_Fake_Data](./assets/mnist_fake.png)

## Fashion-MNIST (fully connected model)

#### Model Architecture:

Generator:

* Input layer: 100 units (Uniform[-1, 1] noise)

* Dense layer: 1200 units, ReLU

* Dense layer: 1200 units, ReLU

* Output layer: 784 units, Sigmoid

* Learnable parameters: 2503984

Discriminator:

* Input layer: 784 units (flattend 28x28 gray-scale image)

* Maxout layer: 240 units, 5 pieces

* Maxout layer: 240 units, 5 pieces

* Output layer: 1 unit, Linear

* Learnable parameters: 1231441

#### Hyperparameters/Settings:

* Epochs: 100  

* Batch size: 128

* Objective: Binary Cross-Entropy Loss with Logits

* Generator optimizer: Adam

* Discriminator optimizer: Adam

* Generator learning rate: 0.0002

* Discriminator learning rate: 0.0002

* Generator betas: (0.5, 0.999)

* Discriminator betas: (0.5, 0.999)

#### Training Data:

![Fashion_MNIST_Training_Data](./assets/fashion_mnist_train.png)

#### GAN Training Losses:

![Fashion_MNIST_Loss](./assets/fashion_mnist_loss.png)

#### Fake Data:

![Fashion_MNIST_Fake_Data](./assets/fashion_mnist_fake.png)

## Citations

```bibtex

@misc{goodfellow2014generativeadversarialnetworks,

      title={Generative Adversarial Networks}, 

      author={Ian J. Goodfellow and Jean Pouget-Abadie and Mehdi Mirza and Bing Xu and David Warde-Farley and Sherjil Ozair and Aaron Courville and Yoshua Bengio},

      year={2014},

      eprint={1406.2661},

      archivePrefix={arXiv},

      primaryClass={stat.ML},

      url={https://arxiv.org/abs/1406.2661}, 

}

```

```bibtex

@misc{goodfellow2013maxoutnetworks,

      title={Maxout Networks}, 

      author={Ian J. Goodfellow and David Warde-Farley and Mehdi Mirza and Aaron Courville and Yoshua Bengio},

      year={2013},

      eprint={1302.4389},

      archivePrefix={arXiv},

      primaryClass={stat.ML},

      url={https://arxiv.org/abs/1302.4389}, 

}

```