CartoonGAN-tensorflow

Simple code implement the paper of CartoonGAN

Introduction

This simple code mainly implement the paper about CartoonGAN, which is published at CVPR2018. This paper mainly address the problem of styling the nature images to cartoon by using GANs.

There are three contributions in the paper:

1. Removing the clear edge of original cartoon images as a new datasets
2. Using high-level feature maps in vgg19 to reconstruct the content
3. Initializing the generator by reconstructing the nature images

Results of the paper

How to use the code

Python packages you need

1. python 3.x
2. tensorflow 1.4.0
3. pillow
4. scipy
5. cv2
6. numpy

Process of using

1. Download the cartoon movie, then using 'vedio2img.py' to extract the cartoon images from the cartoon movie. Finally, put the extracted cartoon imges into the folder 'c'.
2. Using 'remove_clear_edge.py' to remove the extracted cartoon images' clear edge as a new datasets, and then put the unclear edge cartoon images into the folder 'e'
3. Download the nature image datasets(we use MSCOCO here, not Flicker), and then put the unzipped MSCOCO datasets into the folder 'MSCOCO'.

Results of our code

I'm sorry, this code is under training ..........