https://github.com/khanovico/cifar-trgan-torch

CIFAR10 image generation using fully-transformer-GAN with pytorch
https://github.com/khanovico/cifar-trgan-torch

gan image-generation pytorch transformer

Last synced: 19 days ago
JSON representation

CIFAR10 image generation using fully-transformer-GAN with pytorch

• Host: GitHub
• URL: https://github.com/khanovico/cifar-trgan-torch
• Owner: khanovico
• License: mit
• Created: 2024-06-23T18:39:04.000Z (8 months ago)
• Default Branch: main
• Last Pushed: 2024-06-25T12:01:25.000Z (8 months ago)
• Last Synced: 2024-11-25T02:07:36.718Z (3 months ago)
• Topics: gan, image-generation, pytorch, transformer
• Language: Python
• Homepage:
• Size: 61.1 MB
• Stars: 2
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# TrGAN

Generation of CIFAR10 images based on TrGAN (hands-on):

## Dependencies

- Python 3.8

- Tensorfow 2.5

## Usage

### Train

1. Use `--dataset_path=` to specify the dataset path (default builds CIFAR-10 dataset), and `--model_name=` to specify the checkpoint directory name.

```

python train.py --dataset_path= --model_name=

```

### Hparams setting

Adjust hyperparameters in the `hparams.py` file.

### Tensorboard

Run `tensorboard --logdir ./`.

## Examples

- CIFAR-10 training progress

![](images/transgan_samples.gif "TransGAN on CIFAR-10")

## References

Code:

- This model depends on other files that may be licensed under different open source licenses.

- TransGAN uses [Differentiable Augmentation](https://arxiv.org/abs/2006.10738). Under BSD 2-Clause "Simplified" License.

- Small-TransGAN models are instances of the original TransGAN architecture with a smaller number of layers and lower-dimensional embeddings.

Implementation notes:

- Single layer per resolution Generator.

- Orthogonal initializer and 4 heads in both Generator and Discriminator.

- WGAN-GP loss.

- Adam with β1 = 0.0 and β2 = 0.99.

- Noise dimension = 64.

- Batch size = 64

## Licence

MIT