/WaveletSRNet

A pytorch implementation of Paper "Wavelet-srnet: A wavelet-based cnn for multi-scale face super resolution"

Primary LanguagePythonMIT LicenseMIT

WaveletSRNet

A pytorch implementation of Paper "Wavelet-srnet: A wavelet-based cnn for multi-scale face super resolution"

Prerequisites

  • Python 2.7
  • PyTorch

Run

Use the default hyparameters except changing the parameter "upscale" according to the expected upscaling factor(2, 3, 4 for 4, 8, 16 upcaling factors, respectively).

CUDA_VISIBLE_DEVICES=1 python main.py --ngpu=1 --test --start_epoch=0 --test_iter=1000 --batchSize=64 --test_batchSize=32 --nrow=4 --upscale=3 --input_height=128 --output_height=128 --crop_height=128 --lr=2e-4 --nEpochs=500 --cuda

Results

Citation

If you use our codes, please cite the following paper:

@inproceedings{huang2017wavelet,
  title={Wavelet-srnet: A wavelet-based cnn for multi-scale face super resolution},
  author={Huang, Huaibo and He, Ran and Sun, Zhenan and Tan, Tieniu},
  booktitle={IEEE International Conference on Computer Vision},
  pages={1689--1697},    
  year={2017}
}

The released codes are only allowed for non-commercial use.