ghljh/multi-scale-progressive-fusion-network

Multi-Scale Progressive Fusion Network for Single Image Deraining

Multi-Scale Progressive Fusion Network for Single Image Deraining (MSPFN)

This is an implementation of the MSPFN model proposed in the paper (Multi-Scale Progressive Fusion Network for Single Image Deraining) with TensorFlow.

Requirements

• Python 3
• TensorFlow 1.12.0
• OpenCV
• tqdm
• glob
• sys

Motivation

The repetitive samples of rain streaks in a rain image as well as its multi-scale versions (multi-scale pyramid images) may carry complementary information (e.g., similar appearance) to characterize target rain streaks. We explore the multi-scale representation from input image scales and deep neural network representations in a unified framework, and propose a multi-scale progressive fusion network (MSPFN) to exploit the correlated information of rain streaks across scales for single image deraining.

Usage

I. Train the MSPFN model

Dataset Organization Form

If you prepare your own dataset, please follow the following form: |--train_data

|--rainysamples  
    |--file1
            ：  
    |--file2
        :
    |--filen
    
|--clean samples
    |--file1
            ：  
    |--file2
        :
    |--filen

Then you can produce the corresponding '.npy' in the '/train_data/npy' file.

$ python preprocessing.py

Training

Download training dataset ((raw images)Baidu Cloud, (Password:4qnh) (.npy)Baidu Cloud, (Password:gd2s)), or prepare your own dataset like above form.

Run the following commands:

cd ./model
python train_MSPFN.py 

II. Test the MSPFN model

Quick Test With the Raw Model (TEST_MSPFN_M17N1.PY)

Download the pretrained models (Baidu Cloud, (Password:u5v6)) (Google Drive).

Download the commonly used testing rain dataset (R100H, R100L, TEST100, TEST1200, TEST2800) (Google Drive), and the test samples and the labels of joint tasks form (BDD350, COCO350, BDD150) (Baidu Cloud, (Password:0e7o)). In addition, the test results of other competing models can be downloaded from here (TEST1200, TEST100, R100H, R100L).

Run the following commands:

cd ./model/test
python test_MSPFN.py

The deraining results will be in './test/test_data/MSPFN'. We only provide the baseline for comparison. There exists the gap (0.1-0.2db) between the provided model and the reported values in the paper, which originates in the subsequent fine-tuning of hyperparameters, training processes and constraints.

Test the Retraining Model With Your Own Dataset (TEST_MSPFN.PY)

Download the pre-trained models.

Put your dataset in './test/test_data/'.

Run the following commands:

cd ./model/test
python test_MSPFN.py

The deraining results will be in './test/test_data/MSPFN'.

Citation

@InProceedings{Kui_2020_CVPR,
	author = {Jiang, Kui and Wang, Zhongyuan and Yi, Peng and Chen, Chen and Huang, Baojin and Luo, Yimin and Ma, Jiayi and Jiang, Junjun},
	title = {Multi-Scale Progressive Fusion Network for Single Image Deraining},
	booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
	month = {June},
	year = {2020}
}
@ARTICLE{9294056,
  author={K. {Jiang} and Z. {Wang} and P. {Yi} and C. {Chen} and Z. {Han} and T. {Lu} and B. {Huang} and J. {Jiang}},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={Decomposition Makes Better Rain Removal: An Improved Attention-guided Deraining Network}, 
  year={2020},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TCSVT.2020.3044887}}