Pyramid Channel-based Feature Attention Network for Image Dehazing

Codes for Pyramid Channel-based Feature Attention Network for Image Dehazing.

Pyramid Channel-based Feature Attention Network for Image Dehazing

Xiaoqin Zhang, Tao Wang, Jinxin Wang, Guiying Tang, Li Zhao

Published on 2020 Computer Vision and Image Understanding (CVIU)

[Paper] [Project Page]

---

Dependency

• Python >= 3.5
• Pytorch >= 1.1
• Torchvision >= 0.4.2
• Pillow >= 5.1.0
• Numpy >= 1.14.3
• Scipy >= 1.1.0

Dataset make

Make you dataset by: Downloading the ITS training and SOTS testing datasets from RESIDE.

1. Training dataset: Put hazy and clear folers from downloaded ITS in ./data/train/ITS/.
2. Testing dataset: put downloaded SOTS (~1000 images) in ./data/testing/SOTS/.
3. Note: train.txt and val.txt provide the image list for training and testing, respectively.

Pretrained Model

We provide the pre-trained models in the following:

• A model trained on ITS for SOTS-Indoor [Baidu drive].
• A model trained on ITS for SOTS-Outdoor [Baidu drive].

Code Introduction

• train.py and test.py are the codes for training and testing the PCFAN.
• ./datasets/datasets.py is used to load the training and testing datasets.
• ./model/network.py defines the structure of PCFAN.
• ./loss/edg_loss.py defines the proposed Edge loss.
• utils.py contains all utilities used for training and testing the PCFAN.
• ./checkpoints/indoor_haze_best.pth and ./checkpoints/outdoor_haze.pth are the trained weights for indoor and outdoor in SOTS from RESIDE.
• The ./logs/indoor_log.log and ./logs/outdoor_log.log record the core logs.
• The ./logs/run_indoor.log and ./logs/run_outdoor.log record the detailed training logs.
• The testing hazy images are saved in ./results/indoor_results/ and ./results/outdoor_results/, respectively.
• The ./data/ folder stores the training and testing data.

Train

You can train the model for indoor dataset by:

python  train.py  --nEpochs 200 --category indoor 

You can train the model for outdoor dataset by:

python  train.py  --nEpochs 10 --category outdoor 

Test

You can test you model on indoor of SOTS dataset.

python test.py --category indoor 

You can test you model on outdoor of SOTS dataset.

python test.py --category outdoor 

Refenrece:

@article{Zhang2020pyramid,
title = {Pyramid Channel-based Feature Attention Network for image dehazing},
author = {Zhang, Xiaoqin and Wang, Tao and Wang, Jinxin and Tang, Guiying  and Zhao, Li},
journal = {Computer Vision and Image Understanding},
volume = {197-198},
pages = {103003},
year = {2020},
publisher={Elsevier}
}