A auto colorization application made by U-Net. About the paper: https://arxiv.org/abs/1505.04597
For runing this program, you need:
Pytorch: 1.1.0 or newer
matplotlib
Result:
Illustrations --> Edges --> Generated
.
.
.
Usage:
Train:
Run train.py, edit the file to change parameters.
Test:
test.py, for test the whole test set
test_costom.py, test one image
test_folder.py, test a folder
Get the pretrained model: https://drive.google.com/file/d/1vI4ZaK3ZnZtX3aah2E7iHwlnBJ59RYAS/view?usp=sharing
About the dataset:
Illustrations are crawled from https://konachan.net/.
Edges are extractd by a laplacian filter.
You can download the dataset from:
https://drive.google.com/open?id=1elHLWSpr-T2aSpAttYYMbDnj7sj5vs-Q
Examples:
Some codes are from: https://github.com/milesial/Pytorch-UNet