/PFFNet

Solution for NTIRE2018 Image Dehazing Challenge & ACCV2018 Kangfu Mei et al.

Primary LanguagePythonMIT LicenseMIT

PFFNet

Our solution for NTIRE2018 Image Dehazing Challenge (20.549db for Indoor and 20.230db for Outdoor), final results could be refer at NTIRE2018. Futher version is accepted by ACCV2018 https://arxiv.org/pdf/1810.02283.pdf. All pretrained models can be found at Here

Preparation

Using data_argument to enchance the datasets, it will produce below datasets

$ python dara_argument.py --fold_A=IndoorTrainHzay --fold_B=IndoorTrainGT --fold_AB=IndoorTrain 

IndoorTrain
    \data   hazy image
    \label  clear image

Train

Using default parameter to train

python train.py --cuda --gpus=4 --train=/path/to/train --test=/path/to/test --lr=0.0001 --step=1000

Test

python test.py --cuda --checkpoints=/path/to/checkpoint --test=/path/to/testimages

Citation

If you use the code in this repository, please cite our paper:

@inproceedings{mei2018pffn,
  title={Progressive Feature Fusion Network for Realistic Image Dehazing},
  author={Mei, Kangfu and Jiang, Aiwen and Li, Juncheng and  Wang, Mingwen},
  booktitle={Asian Conference on Computer Vision (ACCV)},
  year={2018}
}