Skip to content
No description or website provided.
Branch: master
Clone or download
Latest commit cc05cc4 Jul 3, 2018
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
cmake First commit Jun 13, 2017
densecrf First commit Jun 13, 2017
examples First commit Jun 13, 2017
exper Fix get_next_minibatch Aug 2, 2017
include/caffe
matlab First commit Jun 13, 2017
python
scripts
src First commit Jun 13, 2017
tools First commit Jun 13, 2017
CAFFE_README.md
CMakeLists.txt First commit Jun 13, 2017
CONTRIBUTING.md
CONTRIBUTORS.md
DEEPLAB_V2_README.md First commit Jun 13, 2017
INSTALL.md First commit Jun 13, 2017
LICENSE
Makefile
Makefile.config.example First commit Jun 13, 2017
README.md Update README.md Jul 3, 2018
caffe.cloc First commit Jun 13, 2017

README.md

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{liu2018when,
  author = {Liu, Ding and Wen, Bihan and Liu, Xianming and Wang, Zhangyang and Huang, Thomas S.},
  title = {When Image Denoising Meets High-Level Vision Tasks: A Deep Learning Approach},
  booktitle={IJCAI},
  year = {2018}
  }
You can’t perform that action at this time.