A keras implemention of the paper Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
tensorflow
keras2
numpy
opencv
$ python data.py
Clean patches are extracted from 'data/Train400' and saved in 'data/npy_data'.
$ python main.py
Trained models are saved in 'snapshot'.
$ python main.py --only_test True --pretrain 'path of saved model'
Noisy and denoised images are saved in 'snapshot'.
The average PSNR(dB) results of different methods on the BSD68 dataset.
| Noise Level | BM3D | DnCNN-S | DnCNN-keras |
|---|---|---|---|
| 25 | 28.57 | 29.23 | 29.21 |