The code for Global and Local Alignment Networks for Unpaired Image-to-Image Translation arxiv

Framework: visualization results:

This code was tested with Pytorch 1.7.0, CUDA 10.2, and Python 3.7

• Install Pytoch 1.7.0, torchvision, and other dependencies from http://pytorch.org
• Install python libraries visdom and dominate for visualization

pip install visdom dominate

• Clone this repo:

git clone https://github.com/ygjwd12345/GLANet.git
cd GLANet

Please refer to the original CUT and CycleGAN to download datasets and learn how to create your own datasets.

    sh ./datasets/download_cyclegan_dataset.sh a2b

Available datasets are: apple2orange, summer2winter_yosemite, horse2zebra, monet2photo, cezanne2photo, ukiyoe2photo, vangogh2photo, maps, facades, iphone2dslr_flower, ae_photos

    sh ./datasets/download_pix2pix_dataset.sh xx

Available datasets are night2day, edges2handbags, edges2shoes, facades, maps

The Cityscapes dataset can be downloaded from https://cityscapes-dataset.com. After that, use the script ./datasets/prepare_cityscapes_dataset.py to prepare the dataset.

• Train the single-modal I2I translation model. Please check run.sh. For instance:

python train.py  \
--dataroot ./datasets/summer2winter \
--name summer2winter \
--model sc \
--gpu_ids 0 \
--lambda_spatial 10 \
--lambda_gradient 0 \
--attn_layers 4,7,9 \
--loss_mode cos \
--gan_mode lsgan \
--display_port 8093 \
--direction BtoA \
--patch_size 64

• Test the FID score for all training epochs, please also check run.sh. For instance:

python test_fid.py \
--dataroot ./datasets/horse2zebra \
--checkpoints_dir ./checkpoints \
--name horse2zebra \
--gpu_ids 0 \
--model sc \
--num_test 0

• Test the KID, cityscape score, D&C, LPIPS, please check run_dc_lpips.sh in evaluations folder. For instance:

python PerceptualSimilarity/lpips_2dirs.py -d0 /data2/gyang/TAGAN/results/summer2winter-F64-mixer/test_350/images/real_B -d1 /data2/gyang/TAGAN/results/summer2winter-F64-mixer/test_350/images/fake_B -o ./example_dists.txt --use_gpu
python3 segment.py test -d ./datasets/cityscapes -c 19 --arch drn_d_22 \
    --pretrained ./drn_d_22_cityscapes.pth --phase val --batch-size 1

Our code is developed based on FSeSim and unguided. We also thank pytorch-fid for FID computation, LPIPS for diversity score, and D&C for density and coverage evaluation.