README.md

Anime_GAN

This repository records my GAN models with Anime.

Dependencies

• Python 3.6+
• PyTorch 0.4.0
• numpy 1.14.1, matplotlib 2.2.2, scipy 1.1.0
• imageio 2.3.0
• tqdm 4.24.0
• keras 2.1.2(for WGAN_keras)
• tensorflow 1.2.1(for WGAN_keras)
• cv2 -- pip install opencv-python(for ConditionalGAN)

DCGAN

you need to download the dataset (onedrive, or baiduyun) named faces.zip, then execute the first step——Cloning the repository, and extract and move it in Anime_GAN/DCGAN/.

1. Cloning the repository

$ git clone https://github.com/FangYang970206/Anime_GAN.git
$ cd Anime_GAN/DCGAN/

2. run the code

$ python main.py 

3. 500 epochs result

WGAN

The WGAN implemented by pytorch has bugs. If anyone can find them, i will appreciate you so much! I implement a keras version Anime_WGAN. Like DCGAN, you should download the dataset (onedrive, or baiduyun) named faces.zip(the same as DCGAN), then execute the first step——Cloning the repository, and extract and move it in Anime_GAN/WGAN_keras/faces/.

1. Cloning the repository

$ git clone https://github.com/FangYang970206/Anime_GAN.git
$ cd Anime_GAN/WGAN_keras/

2. run the code

$ python wgan_keras.py 

3. 100000 steps result

ConditionalGAN

you need to download the dataset (onedrive or baiduyun) named images.zip, then execute the first step——Cloning the repository, and extract and move it in Anime_GAN/ConditionalGAN/.The tag.csv file is dataset label, include hair color and eye color.The test.txt is test label file used to generate test images.

1. Cloning the repository

$ git clone https://github.com/FangYang970206/Anime_GAN.git
$ cd Anime_GAN/ConditionalGAN/

2. run the code

Two mode: train or infer. In infer mode, make sure you have trained the ConditionalGAN.

$ python main.py --mode "train"
$ python main.py --mode "infer"  

3. 55000 steps result

4. GUI

Before run GUI.py, you should train the conditionalGAN to get the generate.t7.

$ python GUI.py

Reference

• 李宏毅GAN课程及PPT
• DCGAN paper
• chenyuntc
• JasonYao81000
• tjwei