Awesome

Fusion from Decomposition

This repository is an official implementation of Fusion from Decomposition: A Self-Supervised Decomposition Approach for Image Fusion (ECCV 2022).

Prerequisites

• Linux
• Python 3
• NVIDIA GPU + CUDA cuDNN
• PyTorch 1.9
• torchvision 0.8
• Pillow 8.1
• Opencv 4.4

Getting Started

Installation

• Install python libraries and requests.
• Clone this repo:

git clone https://github.com/erfect2020/DecompositionForFusion.git
cd DecompositionForFusion

Start run

1. Download COCO: https://cocodataset.org/
2. Put your training images into any floder and modify the `option/train/SelfTrained_SDataset.yaml' to retarget the path.
3. Train DeFusion

   python selftrain.py --opt options/train/SelfTrained_SDataset.yaml
   

Start evaluation

1. Download test dataset:
   1. Multi-exposure image fusion: MEFB:https://github.com/xingchenzhang/MEFB, SICE:https://github.com/csjcai/SICE.
   2. Multi-focus image fusion: Real-MFF:https://githubmemory.com/repo/Zancelot/Real-MFF, Dataset:https://github.com/xingchenzhang/MFIFB.
   3. Visible-infrared image fusion: RoadScene:https://github.com/jiayi-ma/RoadScene, TNO:https://figshare.com/articles/dataset/TNO_Image_Fusion_Dataset/1008029.
2. Modify test.py to select the data preprocess files for different tasks:
3. (Option) Our pretrained model is avaliable at Google Drive
4. Test DeFusion
   1. Test multi-exposure image fusion task on MEFB or SICE

   python test.py --opt options/test/MEF_Test_Dataset.yaml or
   python test.py --opt options/test/SMEF_Test_Dataset.yaml
   

   2. Test multi-focus image fusion task on Real-MFF or Dataset

   python test.py --opt options/test/MFF_Test_Dataset.yaml or
   python test.py --opt options/test/EMFF_Test_Dataset.yaml
   

   3. Test visible infrared image fusion task on RoadScene or TNO

   python test.py --opt options/test/IVF_Test_Dataset.yaml or
   python test.py --opt options/test/TIVF_Test_Dataset.yaml
   

License

Distributed under the MIT License. See LICENSE.md for more information.

Citations

If DeFusion helps your research or work, please consider citing DeFusion.

@InProceedings{Liang2022ECCV,
    author    = {Liang, Pengwei and Jiang, Junjun and Liu, Xianming and Ma, Jiayi},
    title     = {Fusion from Decomposition: A Self-Supervised Decomposition Approach for Image Fusion},
    booktitle = {European Conference on Computer Vision (ECCV)},
    year      = {2022},
}