https://github.com/sfikas/quaternion-gan

Code that implements the Quaternion Generative Adversarial Network.
https://github.com/sfikas/quaternion-gan

deep-learning gan generative-adversarial-network pytorch quaternion

Last synced: 22 days ago
JSON representation

Code that implements the Quaternion Generative Adversarial Network.

• Host: GitHub
• URL: https://github.com/sfikas/quaternion-gan
• Owner: sfikas
• Created: 2021-01-29T13:53:53.000Z (about 4 years ago)
• Default Branch: main
• Last Pushed: 2021-12-09T14:25:24.000Z (about 3 years ago)
• Last Synced: 2024-11-14T21:11:58.546Z (3 months ago)
• Topics: deep-learning, gan, generative-adversarial-network, pytorch, quaternion
• Language: Python
• Homepage:
• Size: 399 KB
• Stars: 1
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Paper code for "Quaternion Generative Adversarial Networks for Inscription Detection in Byzantine Monuments"

This repository contains source code for the paper "

[Quaternion Generative Adversarial Networks for Inscription Detection in Byzantine Monuments](http://cs.uoi.gr/~sfikas/icprw-quaternion-gan.pdf)

" by Sfikas et al. (PatReCH 2020).

![QGAN results](figure.png "Sample")

## Requirements

Use ``pip3 install -r requirements.txt`` (after creating a virtual environment optionally, followed by ``pip3 install --upgrade pip``).

Subsequently, please follow first the instructions in https://github.com/Orkis-Research/Pytorch-Quaternion-Neural-Networks to install the Quaternion Convolution layer code.

## Dataset

The data used for the paper may be downloaded from [this google drive link](https://drive.google.com/file/d/1tDjFgLaDxl7gcOw9NPfWjAl3sCNvEhyv/view?usp=sharing).

Unzip the tgz file so that a ```fixtures/bessarion-midi``` folder is created under the repository home directory.

## Code execution

The main executable is ```quaternion-gan.py```, which will train the model using data found in the ```fixtures/``` folder.

You can use the code by supplying your own data, which should come in pairs of image (png) and annotation (npz) files (see BessarionMini class for details).

If you find the paper and/or the code useful, please cite the paper using the following bibtex code:

```

@inproceedings{sfikas2021quaternionGAN,

  title={Quaternion Generative Adversarial Networks for Inscription Detection in Byzantine Monuments},

  author={Sfikas, Giorgos and Giotis, A.P. and Retsinas, George and Nikou, Christophoros},

  booktitle={2^{nd} International Workshop on Pattern Recognition for Cultural Heritage (PatReCH)},

  year={2021}

}

```