The source code of CVPR 2020 paper "Multi-Scale Boosted Dehazing Network with Dense Feature Fusion" by Hang Dong, Jinshan Pan, Zhe Hu, Xiang Lei, Xinyi Zhang, Fei Wang, Ming-Hsuan Yang
- Python 3.6
- PyTorch >= 1.1.0
- torchvision
- numpy
- skimage
- h5py
- MATLAB
-
Download the Pretrained model on RESIDE and Test set to
MSBDN-DFF/models
andMSBDN-DFF/
folder, respectively. -
Run the
MSBDN-DFF/test.py
with cuda on command line:
MSBDN-DFF/$python test.py --checkpoint path_to_pretrained_model
- The dehazed images will be saved in the directory of the test set.
We find the choices of training images play an important role during the training stage, so we offer the training set of HDF5 format:
Baidu Yun (code:v8ku)
You can use the DataSet_HDF5() in ./datasets/dataset_hf5.py to load these HDF5 files.
If you use these models in your research, please cite:
@conference{MSBDN-DFF,
author = {Hang, Dong and Jinshan, Pan and Zhe, Hu and Xiang, Lei and Fei, Wang and Ming-Hsuan, Yang},
title = {Multi-Scale Boosted Dehazing Network with Dense Feature Fusion},
booktitle = {CVPR},
year = {2020}
}