Paper | Bibtex

The code is built on DnCNN (PyTorch) and tested on Ubuntu 16.04/18.04 environment (Python3.6, PyTorch_1.3.0, CUDA10.1, cuDNN7.6) with Titan RTX/2080Ti GPUs.

python3
pytorch >= 1.3
torchvision >= 0.4.1.
tensorboardX
hdf5
numpy
opencv

$ python train.py --preprocess True

NOTE

• If you've already built the training and validation dataset (i.e. train.h5 & val.h5 files), set preprocess to be False.

$ python test.py --test_data Set12 --output_size 10 --test_noiseL 25

NOTE

• test_data can be Set12 or Set68 or Urban100.
• output_size can be any number from 1. Please refer to the paper on this part.
• test_noiseL can be 15 or 25 or 50
• pretrained weights for each noise level are provided in logs/

@article{lee2020locally,
  title={Locally Adaptive Channel Attention-Based Network for Denoising Images},
  author={Lee, Haeyun and Cho, Sunghyun},
  journal={IEEE Access},
  volume={8},
  pages={34686--34695},
  year={2020},
  publisher={IEEE}
}