Awesome

pytorch-CartoonGAN

Pytorch implementation of CartoonGAN [1] (CVPR 2018)

Parameters without information in the paper were set arbitrarily.
I used face-cropped celebA (src) and anime (tgt) collected from the web data because I could not find the author's data.

Tensorflow version

CartoonGAN-tensorflow

Usage

1.Download VGG19

VGG19

2.Train

python CartoonGAN.py --name your_project_name --src_data src_data_path --tgt_data tgt_data_path --vgg_model pre_trained_VGG19_model_path

Folder structure

The following shows basic folder structure.

├── data
│   ├── src_data # src data (not included in this repo)
│   │   ├── train 
│   │   └── test
│   └── tgt_data # tgt data (not included in this repo)
│       ├── train 
│       └── pair # edge-promoting results to be saved here
│
├── CartoonGAN.py # training code
├── edge_promoting.py
├── utils.py
├── networks.py
└── name_results # results to be saved here

Resutls

paper results

celebA2anime face

Initialization phase (reconstruction)

<table align='center'> <tr align='center'> <td> Input - Result (this repo) </td> </tr> <tr> <td><img src = 'assets/Initialization_phase_result1.png'> </tr> <tr> <td><img src = 'assets/Initialization_phase_result2.png'> </tr> <tr> <td><img src = 'assets/Initialization_phase_result3.png'> </tr> </table>

Cartoonization

I got the author's results from CaroonGAN-Test-Pytorch-Torch.

<table align='center'> <tr align='center'> <td> Input - Result (this repo) </td> <td> Author's pre-trained model (Hayao) </td> <td> Author's pre-trained model (Hosoda) </td> </tr> <tr> <td><img src = 'assets/Cartoonization_result1.png'> <td><img src = 'assets/00055_Hayao.jpg' height = '200px'> <td><img src = 'assets/00055_Hosoda.jpg' height = '200px'> </tr> <tr> <td><img src = 'assets/Cartoonization_result2.png'> <td><img src = 'assets/00009_Hayao.jpg' height = '200px'> <td><img src = 'assets/00009_Hosoda.jpg' height = '200px'> </tr> <tr> <td><img src = 'assets/Cartoonization_result3.png'> <td><img src = 'assets/00010_Hayao.jpg' height = '200px'> <td><img src = 'assets/00010_Hosoda.jpg' height = '200px'> </tr> <tr> <td><img src = 'assets/Cartoonization_result4.png'> <td><img src = 'assets/00155_Hayao.jpg' height = '200px'> <td><img src = 'assets/00155_Hosoda.jpg' height = '200px'> </tr> <tr> <td><img src = 'assets/Cartoonization_result5.png'> <td><img src = 'assets/00014_Hayao.jpg' height = '200px'> <td><img src = 'assets/00014_Hosoda.jpg' height = '200px'> </tr> </table>

Development Environment

NVIDIA GTX 1080 ti
cuda 8.0
python 3.5.3
pytorch 0.4.0
torchvision 0.2.1
opencv 3.2.0

Reference

[1] Chen, Yang, Yu-Kun Lai, and Yong-Jin Liu. "CartoonGAN: Generative Adversarial Networks for Photo Cartoonization." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018.

(Full paper: http://openaccess.thecvf.com/content_cvpr_2018/papers/Chen_CartoonGAN_Generative_Adversarial_CVPR_2018_paper.pdf)