Unsupervised-end-to-end-infrared-and-visible-image-fusion-network-using-learnable-fusion-strategy

This repository contains the codes for paper Unsupervised end-to-end infrared and visible image fusion network using learnable fusion strategy by Yili Chen, Minjie Wan*, Yunkai Xu, et al. (*Corresponding author).

The overall repository style is partially borrowed from PFNet (https://github.com/Junchao2018/Polarization-image-fusion). Thanks to Junchao Zhang.

The visible-thermal dataset TNO can be downloaded from https://figshare.com/articles/dataset/TNO_Image_Fusion_Dataset/1008029, the multi-focus dataset Lytro can be downloaded from https://mansournejati.ece.iut.ac.ir/content/lytro-multi-focus-dataset.

Requirements

Python==3.7

Tensorflow==1.13.1

cuda==10.0.13 and cudnn

h5py

opencv-python

other packages if needed

Usage

1. Train and test data using GenerateTrainingPatches_Tensorflow.m and GenerateTestingPatches_Tensorflow.m, please create two folders named TrainingData and TestingData and put the generated mats into them respectively.
2. Train your own model using backward.py, and the relative parameters can be adjusted in the same file.
3. Generate fusion results with the trained model by test.py.
4. The default output format is '.mat', and you can use Matlab to convert it to other common figure formats.