Pytorch implementation of the paper:
Artistic Photo Filter Removal Using CNNs
Journal of Electronic Imaging, SPIE
F. Piccoli, C. Cusano, S. Bianco, R. Schettini
Usage:
# clone this repository
git clone --recursive https://github.com/dros1986/filter_removal.git
# download the dataset
wget https://drive.google.com/a/campus.unimib.it/uc?export=download&confirm=XAOn&id=1vvLAO__opCjgLfRjAjW3WPWJHNiiVLbs
# unzip the file
unzip file.zip -d ./datasets/
# start training
python main.py -degin 3 degout 3
# start test
python main.py -degin 3 degout 3 --regen ./checkpoint.pthInput images
Output images
| Name | Description | Default |
|---|---|---|
| degin | Degree of the polynomial onto which the color transform will be estimated | 3 |
| degout | Degree of the polynomial onto which the color transform will be applied | 3 |
| patchsize | patchsize*patchsize is the number of pixels involved in each color transform | 8 |
| nrow | Batch size will be nrow*nrow | 5 |
| indir | Folder containing filtered images | ./datasets/places-instagram/images/ |
| gtdir | Folder containing original images | ./datasets/places-instagram/images_orig/ |
| train_list | txt containing train set filenames | ./datasets/places-instagram/train-list.txt |
| validation_list | txt containing validation set filenames | ./datasets/places-instagram/smallvalidation-list.txt |
| test_list | txt containing test set filenames | ./datasets/places-instagram/test-list.txt |