MingtaoGuo / SinGAN_Pytorch

Reimplementing the paper "SinGAN: Learning a Generative Model from a Single Natural Image"

SinGAN_Pytorch

Reimplementing the paper "SinGAN: Learning a Generative Model from a Single Natural Image"

Requirements

1. python3
2. torch1.1.0
3. pillow
4. numpy
5. imageio

Training phase what you need to do

1. Modifying the image path in "train.py"
2. Executing the file "train.py"

Testing phase what you can do

1. Random sample from single image, "random_sample_from_single.py"
2. Harmonization, "harmonization.py"
3. Creating an animation, "animation.py"
4. Converting painting to image, "harmonization.py"

Results

Raw img	random sampled	animation

Harmonization

Raw img	n=1	n=2	n=3	n=4

Raw img	n=5	n=6	n=7	n=8

Problems

The results of this code still have some problems. Sometimes, it generates the distortion image. I really don't know how to fix it.

Acknowledgement

Thanks for the source code of SinGAN, it's very helpful!

Author

Mingtao Guo

Xi'an University of technology

Reference

[1]. Shaham, Tamar Rott, Tali Dekel, and Tomer Michaeli. "Singan: Learning a generative model from a single natural image." Proceedings of the IEEE International Conference on Computer Vision. 2019.