# anime-StyleGAN

**Repository Path**: MorvanZhou/anime-StyleGAN

## Basic Information

- **Project Name**: anime-StyleGAN
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2021-01-15
- **Last Updated**: 2021-01-25

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

An implementation of StyleGAN on Animation dataset.

## Install 

```shell script
git clone https://github.com/MorvanZhou/anime-StyleGAN
cd anime-StyleGAN
pip install -r requirements.txt
```

## Process data
Download data from kaggle [Anime Faces](https://www.kaggle.com/soumikrakshit/anime-faces) (~400MB), 
then unzip \*.zip to the anime/images folder.

process data to tensorflow tensor_record format.

```shell script
python dataset.py --data_dir ~/data/anime/
```

## Train

```shell script
python train.py --data_dir ~/data/anime/ --soft_gpu -b 32 --latent 128 --epoch 151 -lr 0.0005 -b1 0. -b2 0.99 -w 2
```

## Results

First few of epoch:

![img](demo/ep000t001000.png)

After one day:

![img](demo/ep141t001000.png)

## Style mixing

```shell script
python restore.py --model_path visual/2021-01-14_19-39-20/model/cp-0137.ckpt -o demo/stylemix.png -n 3 --z_seed 2 --n_seed 1 --latent 128
```

![img](demo/stylemix.png)