Official Pytorch implementation of "Patch-wise Graph Contrastive Learning for Image Translation" (AAAI 2024)

Chanyong Jung, Gihyun Kwon, Jong Chul Ye

Link: https://arxiv.org/abs/2312.08223

• We provide the pretrained model: horse-to-zebra
• For the training from scratch (ex. horse2zebra dataset):

  python train.py --name [folder-name] --dataroot [path-to-data] \
  --lambda_GNN 0.1 --num_hop 2 --gnn_idt --nonzero_th 0.1 \
  --pooling_num 1 --pooling_ratio '1,0.5' --down_scale 4 \
  --gpu_ids 0

• For evaluation:

  
  python test.py --dataroot [path-to-dataset] --name [experiment-name] \
  --CUT_mode CUT --phase test --epoch [epoch-for-test] --num_test [test-size]

python -m pytorch_fid [path-to-output] [path-to-input]

### Cite

@article{jung2023patch, title={Patch-wise Graph Contrastive Learning for Image Translation}, author={Jung, Chanyong and Kwon, Gihyun and Ye, Jong Chul}, journal={arXiv preprint arXiv:2312.08223}, year={2023} }

### Acknowledgement
Our source code is based on the official implementation of [HnegSRC](https://github.com/jcy132/Hneg_SRC) and [CUT](https://github.com/taesungp/contrastive-unpaired-translation).