Official repository of "Deep Coupled Feedback Network for Joint Exposure Fusion and Image Super-Resolution"
The training data and testing data is from the [SICE dataset]. Or you can download the datasets from our [Google Drive Link].
git clone https://github.com/ytZhang99/CF-Net.git
dataset/test_data/lr_over
and dataset/test_data/lr_under
, respectively.
dataset
└── test_data
├── lr_over
└── lr_under
python main.py --test_only --scale 2 --model model_x2.pth
python main.py --test_only --scale 4 --model model_x4.pth
./test_results
.dataset/val_data
.
dataset
├── train_data
| ├── hr
| ├── hr_over
| └── hr_under
└── val_data
├── gt
├── lr_over
└── lr_under
Prepare_Data_HR_LR.m
file and modify the following lines according to your training commands.
Line 5 or 6 : scale = 2 or 4
Line 9 : whether use off-line data augmentation (default = True)
[Line 12 <-> Line 17] or [Line 13 <-> Line 18] : producing [lr_over/lr_under] images from [hr_over/hr_under] images
dataset/train_data
should be as follows:
dataset
└── train_data
├── hr
├── hr_over
├── hr_under
├── lr_over
└── lr_under
dataset.py
and train.py
in the same directory with main.py
.python main.py --scale 2 --model my_model
python main.py --scale 4 --model my_model
If validation data is added, run the following command to get the best model best_ep.pth
.
python main.py --scale 2 --model my_model -v
python main.py --scale 4 --model my_model -v
./model/
.If you find our work useful in your research or publication, please cite our work:
@article{deng2021deep,
title={Deep Coupled Feedback Network for Joint Exposure Fusion and Image Super-Resolution.},
author={Deng, Xin and Zhang, Yutong and Xu, Mai and Gu, Shuhang and Duan, Yiping},
journal={IEEE Transactions on Image Processing: a Publication of the IEEE Signal Processing Society},
year={2021}
}
If you have any question about our work or code, please email [email protected]
.