Unsupervised Face Super-Resolution via Gradient Enhancement and Semantic Guidance (GESGNet)

Preparation

prepare unpaired data, which includes images of HR domain and LR domain
use ESRGAN to transform LR domain into inter-HR domain
put images of inter-HR and HR domain in a folder your_datasets, and rename them as trainA and trainB, respectively.
download the pretrained face parsing model, and put it in the folder models. (The original face parsing codes can be found in https://github.com/zllrunning/face-parsing.PyTorch)

Training

python train.py --name experiment_name --dataroot /path/to/your_datasets --model GESG

Testing

python test.py --name experiment_name --dataroot /path/to/your_datasets --model GESG