felhe / ridnet

Pytorch code for "Real image denoising with feature attention", ICCV (Oral), 2019.

Real Image Denoising with Feature Attention

This repository is for Real Image Denoising with Feature Attention (RIDNet) introduced in the following paper

Saeed Anwar, [Nick Barnes], "Real Image Denoising with Feature Attention", ICCV (Oral), 2019

The model is built in PyTorch 0.4.0, PyTorch 0.4.1 and tested on Ubuntu 14.04/16.04 environment (Python3.6, CUDA9.0, cuDNN5.1).

Contents

1. Introduction
2. Network
3. Train
4. Test
5. Results
6. Citation
7. Acknowledgements

Introduction

Deep convolutional neural networks perform better on images containing spatially invariant noise (synthetic noise); however, their performance is limited on real-noisy photographs and requires multiple stage network modeling. To advance the practicability of denoising algorithms, this paper proposes a novel single-stage blind real image denoising network (RIDNet) by employing a modular architecture. We use a residual on the residual structure to ease the flow of low-frequency information and apply feature attention to exploit the channel dependencies. Furthermore, the evaluation in terms of quantitative metrics and visual quality on three synthetic and four real noisy datasets against 19 state-of-the-art algorithms demonstrate the superiority of our RIDNet.

Sample results on a real noisy face image from RNI15 dataset.

Network

The architecture of the proposed network. Different green colors of the conv layers denote different dilations while the smaller size of the conv layer means the kernel is 1x1. The second row shows the architecture of each EAM.

The feature attention mechanism for selecting the essential features.

Train

Will be added later

Test

Quick start

1. Download the trained models for our paper and place them in '/TestCode/experiment'.

   The real denoising model can be downloaded from Google Drive or here. The total size for all models is 5MB.

2. Cd to '/TestCode/code', run the following scripts.

   You can use the following script to test the algorithm

   #RIDNET
   CUDA_VISIBLE_DEVICES=0 python main.py --data_test MyImage --noise_g 1 --model RIDNET --n_feats 64 --pre_train ../experiment/ridnet.pt --test_only --save_results --save 'RIDNET_RNI15' --testpath ../LR/LRBI/ --testset RNI15

Results

All the results for RIDNET can be downloaded from GoogleDrive from SSID, RNI15 and DnD. The size of the results is 65MB

Quantitative Results

The performance of state-of-the-art algorithms on widely used publicly available DnD dataset in terms of PSNR (in dB) and SSIM. The best results are highlighted in bold.

The quantitative results (in PSNR (dB)) for the SSID and Nam datasets.. The best results are presented in bold.

For more information, please refer to our paper

Visual Results

A real noisy example from DND dataset for comparison of our method against the state-of-the-art algorithms.

Comparison on more samples from DnD. The sharpness of the edges on the objects and textures restored by our method is the best.

A real high noise example from RNI15 dataset. Our method is able to remove the noise in textured and smooth areas without introducing artifacts

A challenging example from SSID dataset. Our method can remove noise and restore true colors

Few more examples from SSID dataset.

Citation

If you find the code helpful in your resarch or work, please cite the following papers.

@article{anwar2019ridnet,
  title={Real Image Denoising with Feature Attention},
  author={Anwar, Saeed and Barnes, Nick},
  journal={IEEE International Conference on Computer Vision (ICCV-Oral)},
  year={2019}
}

@article{Anwar2020IERD,
  author = {Anwar, Saeed and Huynh, Cong P. and Porikli, Fatih },
    title = {Identity Enhanced Image Denoising},
    journal={IEEE Computer Vision and Pattern Recognition Workshops (CVPRW)},
    year={2020}
}

Acknowledgements

This code is built on DRLN (PyTorch)