MFF-GAN

Code of paper MFF-GAN: An unsupervised generative adversarial network with adaptive and gradient joint constraints for multi-focus image fusion.

@article{zhang2021mff,
  title={MFF-GAN: An unsupervised generative adversarial network with adaptive and gradient joint constraints for multi-focus image fusion},
  author={Zhang, Hao and Le, Zhuliang and Shao, Zhenfeng and Xu, Han and Ma, Jiayi},
  journal={Information Fusion},
  volume={66},
  pages={40--53},
  year={2021},
  publisher={Elsevier}
}

Recommended Environment:

• python = 2.7
• tensorflow-gpu = 1.9.0
• numpy = 1.15.4
• h5py = 2.9.0
• scipy = 1.2.0
• opencv = 2.4.11

Prepare data :

Run "main.m" (the first function) to convert source images from RGB color space to YCbCr.

To train :

Put training image pairs (Y channel) in the "Train_near" and "Train_far" folders, and run "CUDA_VISIBLE_DEVICES=0 python main.py" to train the network.

To test :

Put test image pairs (Y channel) in the "Test_near" and "Test_far" folders, and run "CUDA_VISIBLE_DEVICES=0 python test.py" to test the trained model. You can also directly use the trained model we provide.

Restore the output of networks to RGB space :

Run "main.m" (the second function) to restore the output of networks to RGB color space.