We provide PyTorch implementation for:
Adversarial Segmentation Loss for Sketch Colorization
AdvSegLoss:
Visual comparison with other methods:
FID score comparison with other methods:
- Linux, macOS or Windows
- Python 3.7
- CPU or NVIDIA GPU + CUDA CuDNN
Please refer to datasets.md for details.
- Clone this repo:
git clone https://github.com/giddyyupp/AdvSegLoss.git
cd AdvSegLoss
- Install PyTorch 1.7.0 and torchvision from https://pytorch.org/get-started/locally.
conda install pytorch=1.7.0 torchvision cudatoolkit=11.0 -c pytorch
Or install all deps using pip:
pip install -r requirements.txt
- Download a dataset (e.g. bedroom) and generate edge maps:
Follow steps in the datasets.md to download, and prepare datasets.
- Train a model with unpaired training:
#!./scripts/train_advsegloss.sh
python train.py --name ade20k_hed_advsegloss_both --dataroot ./datasets/ade20k_hed --model cycle_gan --segmentation --segmentation_output "both" --direction "AtoB" --dataset_mode "unaligned"
- To view training results and loss plots, run
python -m visdom.server
and click the URL http://localhost:8097. To see more intermediate results, check out./checkpoints/maps_cyclegan/web/index.html
- Test the model:
#!./scripts/test_advsegloss.sh
python test.py --name ade20k_hed_advsegloss_both --dataroot ./datasets/ade20k_hed --model test --segmentation --segmentation_output "both" --direction "AtoB" --dataset_mode "unaligned"
The test results will be saved to a html file here: ./results/ade20k_hed_advsegloss_both/latest_test/index.html
.
You can find more scripts at scripts
directory.
- You can download pretrained models using following link
Put a pretrained model under ./checkpoints/{name}_pretrained/100_net_G.pth
.
- Then generate the results using
python test.py --dataroot datasets/ade20k_hed/testB --name {name}_pretrained --model test --segmentation --segmentation_output "both" --direction "AtoB" --dataset_mode "unaligned"
The results will be saved at ./results/
. Use --results_dir {directory_path_to_save_result}
to specify the results directory.
Best practice for training and testing your models.
Before you post a new question, please first look at the above Q & A and existing GitHub issues.
Our code is based on Ganilla.
The numerical calculations reported in this work were fully performed at TUBITAK ULAKBIM, High Performance and Grid Computing Center (TRUBA resources).