uzeful/IFCNN

code for "IFCNN: A General Image Fusion Framework Based on Convolutional Neural Network"

IFCNN

Project page of "IFCNN: A General Image Fusion Framework Based on Convolutional Neural Network, Information Fusion, 54 (2020) 99-118".

Requirements

• pytorch=0.4.1
• python=3.x
• torchvision
• numpy
• opencv-python
• jupyter notebook (optional)
• anaconda (suggeted)

Configuration

# Create your virtual environment using anaconda
conda create -n IFCNN python=3.5

# Activate your virtual environment
conda activate IFCNN

# Install the required libraries
conda install pytorch=0.4.1 cuda80 -c pytorch
conda install torchvision numpy jupyter notebook
pip install opencv-python

Usage

# Clone our code
git clone https://github.com/uzeful/IFCNN.git
cd IFCNN/Code

# Remember to activate your virtual enviroment before running our code
conda activate IFCNN

# Replicate our image method on fusing multiple types of images
python IFCNN_Main.py

# Or run code part by part in notebook
jupyter notebook IFCNN_Notebook.ipynb

Typos

1. Eq. (4) in our paper is wrongly written, the correct expression can be referred to the official expression in OpenCV document, i.e., , where , , , and is the scale factor chosen for achieving .
2. Stride and padding parameters of CONV4 are respectively 1 and 0, rather than both 0.

Highlights

• Propose a general image fusion framework based on convolutional neural network
• Demonstrate good generalization ability for fusing various types of images
• Perform comparably or even better than other algorithms on four image datasets
• Create a large-scale and diverse multi-focus image dataset for training CNN models
• Incorporate perceptual loss to boost the model’s performance

Architecture of our image fusion model

Comparison Examples

1. Multi-focus image fusion

2. Infrared and visual image fusion

3. Multi-modal medical image fusion

4. Multi-exposure image fusion

Other Results of Our Model

1. Multi-focus image dataset: Results/CMF
2. Infrared and visual image dataset: Results/IV
3. Multi-modal medical image dataset: Results/MD
4. Multi-exposure image dataset: Results/ME

Citation

If you find this code is useful for your research, please consider to cite our paper. Yu Zhang, Yu Liu, Peng Sun, Han Yan, Xiaolin Zhao, Li Zhang, IFCNN: A General Image Fusion Framework Based on Convolutional Neural Network, Information Fusion, 54 (2020) 99-118.

@article{zhang2020IFCNN,
  title={IFCNN: A General Image Fusion Framework Based on Convolutional Neural Network},
  author={Zhang, Yu and Liu, Yu and Sun, Peng and Yan, Han and Zhao, Xiaolin and Zhang, Li},
  journal={Information Fusion},
  volume={54},
  pages={99--118},
  year={2020},
  publisher={Elsevier}
}