DCGAN

Introduction

Implement DCGAN from scratch using Pytorch.

Usage

1. Clone this repository

git clone https://github.com/tien02/Implement-DCGAN.git

2. Create virtual environment (venv), then activate it.

python -m venv your_venv

3. Install dependencies

pip install -r requirements.txt

4. Run training process, use --help for more details

python train.py [-h]
                [--datapath] path/to/your/dataset
                [--checkpoint] path/to/your/checkpoint
                [--save] 
                [--tensorboard]
                [--make_gif]

Use --save to save checkpoint after training.

Use --tensorboard to activate tensorboard in order to see the training process.

Use --make_gif to make create .gif image file about generation process after training.

Result

• Result on MNIST dataset:

Notice

• Edit config.py for your own purpose.

• GAN is really sensitive to hyperparameters, be hesitate before tuning.

References

• Paper:
  • Generative Adversarial Nets
  • Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• YouTube: Aladdin Persson