yyfyan / FAMED-Net

FAMED-Net: A Fast and Accurate Multi-scale End-to-end Dehazing Network

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

FAMED-Net

FAMED-Net: A Fast and Accurate Multi-scale End-to-end Dehazing Network.

The code has been tested on Ubuntu 14.04 with CUDA 8.0.

Installation

Install the caffe-master-FAMED-Net and compile the Matlab interface.

If you use Ubuntu 16.04, please modify Makefile and Makefile.config.

Download the AOD-Net model and FPC-Net model from AOD-Net and FPC-Net, rename them as "AOD_Net.caffemodel" and "FPC-Net.caffemodel", and put them into the "model" folder.

Folder Structure

caffe-master-FAMED-Net
    The caffe source code
FAMED-Net
    -fast-guided-filter
        Fast guided filter code [1]
    -generateData
        Generating HDF5 training files
    -model
        Folder containing dehazed models of AOD-Net [2], FPC-Net [3], and FAMED-Net
    -results
        Folder containing dehazed results
    -stats
        Codes and data for generating the learned statistical priors of different models
    -testImgs
        Test hazy images
    -utils
        PSNR, SSIM (from [2]), and store2hdf5 functions
    -testDemoObjectiveEval_ForTestSet_FastGF
        Main script for objective evaluation on RESIDE SOTS test set
    -testDemoSubjectiveEval_ForImgs_FastGF
        Main script for subjective evaluation on single hazy test image

Reference

[1]. Fast guided filter, FGF

[2]. A Benchmark for Single Image Dehazing, RESIDE

[3]. Fully Point-wise Convolutional Neural Network for Modeling Statistical Regularities in Natural Images, FPC-Net

Citation

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

@article{zhang2019famednet, 
    author={Zhang, Jing and Tao, Dacheng}, 
    journal={IEEE Transactions on Image Processing}, 
    title={FAMED-Net: A Fast and Accurate Multi-scale End-to-end Dehazing Network}, 
    year={2019}, 
    volume={}, 
    number={}, 
    pages={1-1}, 
    doi={10.1109/TIP.2019.2922837}, 
    ISSN={1057-7149}, 
    month={}
}

Related Work

[1]. Fully Point-wise Convolutional Neural Network for Modeling Statistical Regularities in Natural Images. FPC-Net: Project, FPC-Net: github

@inproceedings{zhang2018fpcnet,
  title={Fully Point-wise Convolutional Neural Network for Modeling Statistical Regularities in Natural Images},
  author={Zhang, Jing and Cao, Yang and Wang, Yang and Wen, Chenglin and Chen, Chang Wen},
  booktitle={ACM Multimedia Conference},
  year={2018}
}

Contact

Email

About

FAMED-Net: A Fast and Accurate Multi-scale End-to-end Dehazing Network


Languages

Language:C++ 78.4%Language:Python 8.4%Language:Cuda 6.9%Language:CMake 2.8%Language:MATLAB 2.1%Language:Makefile 0.6%Language:Shell 0.3%Language:CSS 0.2%Language:Dockerfile 0.1%Language:HTML 0.1%