Skip to content

wenbihan/DeepDenoising

 
 

Repository files navigation

DeepDenoising

Introduction

This repository provides codes that we use to study the mutual influence between image denoising and high-level vision tasks.

(1) We present an image denoising network which achieves state-of-the-art image denoising performance.

(2) We propose a deep network solution that cascades two modules for image denoising and various high-level tasks, respectively, and demonstrate that the proposed architecture not only yields superior image denoising results preserving fine details, but also overcomes the performance degradation of different high-level vision tasks, such as image classification and semantic segmentation, due to image noise.

This code repository is built on top of DeepLab v2.

For more details, please refer to our paper.

Download models

  • cd exper/model/
  • Run get_models.sh to download models used in our work.

Training

  • cd exper
  • Run main_train_denoise.sh to train the denoising network seperately.
  • Run main_train_cls.sh to jointly train the cascade of the denoising network and the network for image classification.
  • Run main_train_seg.sh to jointly train the cascade of the denoising network and the entwork for semantic segmentation.

Testing

  • cd exper
  • Run main_test_cls.sh to test the resulting model for image classification.
  • Run main_test_seg.sh to test the resulting model for semantic segmentation.
  • Run main_test_denoise.sh to generate denoised results.

Citation

Please cite the paper in your publications if it helps your research:

@inproceedings{liu2017when,
  author = {Liu, Ding and Wen, Bihan and Liu, Xianming and Huang, Thomas S.},
  title = {When Image Denoising Meets High-Level Vision Tasks: A Deep Learning Approach},
  year = {2017},
  journal={arXiv preprint arXiv:1706.04284}
  }

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C++ 77.5%
  • Python 8.1%
  • Cuda 6.2%
  • MATLAB 3.2%
  • CMake 2.5%
  • Protocol Buffer 1.5%
  • Other 1.0%