Introduce and explain code; provide steps for implementation.

# Reproducibility Guide for Applying GANs and Diffusion Models in Font Design for Ethnic Minority Writing Systems

## Introduction
To ensure reproducibility,  the entire process involves the following steps:
1.Prepare the environment: set up training environment according to the requirements of models which will be used.
2.Data preprocessing: Use our code to build a dataset.
3.Train the models and use these pre-trained GANs models to generate glyph images. 
4.Use DM model to optimize the glyph images generated by the pre-trained GANs models.


## Requirements
The GANs models used in this study include：
[font_translator_gan]: https://github.com/ligoudaner377/font_translator_gan
[MF-Net ]: https://github.com/iamyufan/MF-Net
[FCAGAN]: https://github.com/jtlxlf/FCAGAN
These three models are reproduced and the new inference is carried out.
Ensure you have the followed requirements installed. It is recommended to use a virtual environment:
- NVIDIA GPU(ensure  CUDA and cuDNN have been installed)
- Python 3
- torch>=0.4.1
- torchvision>=0.2.1
- dominate>=2.3.1
- visdom>=0.1.8.3
copy code
conda create -n glyph_env python=3.8 -y
conda activate glyph_env
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
pip install dominate>=2.3.1 visdom>=0.1.8.3


## Data
Place your glyph images dataset in ./data/ (e.g., handwritten or printed fonts).
Or use the data provided by font_translator_gan (https://github.com/ligoudaner377/font_translator_gan)：https://drive.google.com/file/d/1XJppxR00pyk5xG-64Ia_BF12XSxeZgfa/view?pli=1
Or use our dataset（MEG-1.0）：DOI 10.6084/m9.figshare.28643645
Or use our code to convert font Ethnic Minority Writing Systems file to glyph images.
the code can be downloaded at：https://colab.research.google.com/drive/1942LOkBzmWVIEvsXywZ2E4-PuWU3qfh9?usp=drive_link


## Application
1.To train a GAN model:
orient: MF-Net (https://github.com/iamyufan/MF-Net) 
copy code
python train.py --dataroot ./datasets/font --model mfnet --name mfnet_train --dataset_mode font --no_dropout --gpu_ids 0 --batch_size 100 --shuffle_dataset

For pre-trained weights, download and place them in ./checkpoints/.

2.To test the pre-trained model
copy code
python test.py --dataroot datasets\font --model mfnet --dataset_mode font --eval --no_dropout --epoch 20 --name mfnet_train 

3.To enhance GAN outputs with a Diffusion Model:
The checkpoint model can be downloaded in: https://civitai.com/models?tag=base+model
The checkpoint model can be find in https://www.vegaai.net/imageToImage


## Results
The results include pre-trained models, glyph images generated using GANs and DM.