[Information Fusion 2024] Diff-IF: Multi-modality image fusion via diffusion model with fusion knowledge prior
Yi, Xunpeng, et al. "Diff-IF: Multi-modality image fusion via diffusion model with fusion knowledge prior." Information Fusion (2024): 102450.
- Create Conda Environment
conda create -n diffif_env python=3.8
conda activate diffif_env
- Install Dependencies
conda install pytorch==2.0.0 torchvision==0.15.0 pytorch-cuda=11.8 -c pytorch -c nvidia
pip install -r requirements.txt
You can also refer to MFNet, RoadScene, LLVIP to prepare your data.
If you want to test only, you should list your dataset as the followed rule:
# Infrared and visible image fusion:
dataset/
your_dataset/
test/
Infrared/
Visible/
# Medical image fusion:
dataset/
your_dataset/
test/
CT-PET-SPECT/
MRI/
We provide the pretrain weights for infrared and visible image fusion and medical image fusion. Download the weight and put it into the weights
folder.
The pretrain weight for infrared and visible image fusion is at Google Drive | Baidu Drive (code: 82nm).
The pretrain weight for medical image fusion is at Google Drive | Baidu Drive (code: 7u1g).
For infrared and visible image fusion or medical image fusion test, you can use:
# Infrared and visible fusion
CUDA_VISIBLE_DEVICES=0 python infer_ddim.py --config config/diff-if-ivf_val.json
# Medical image fusion
CUDA_VISIBLE_DEVICES=0 python infer_ddim.py --config config/diff-if-mif_val.json
Please refer to existing methods to achieve fusion knowledge prior construction of training sets. We recommend the U2Fusion, TarDAL, DDFM, MetaFusion, etc. You can also organize your own fusion knowledge prior based on your needs. We encourage the researchers to do this.
You should list your fusion knowledge prior as followed rule:
dataset/
fusion_knowledge_prior/
Knowledge_U2Fusion/
Knowledge_TarDAL/
Knowledge_DDFM/
Knowledge_MetaFusion/
...
We also encourage the researchers to use the customized targeted search based on the needs.
python targeted_search.py
Please move your fusion knowledge (Fusion_K) into the training datasets before training the model. For infrared and visible image fusion, you can list your dataset as followed rule (the folder of Fusion_K will be produced by the targeted search):
dataset/
your_dataset/
train/
Fusion_K/
Infrared/
Visible/
eval/
Infrared/
Visible/
Similarly, the way of training the medical image fusion is simple, referring to the section of prepared your dataset.
# Infrared and visible fusion
CUDA_VISIBLE_DEVICES=0 python train.py --config config/diff-if-ivf.json
# Medical image fusion
CUDA_VISIBLE_DEVICES=0 python train.py --config config/diff-if-mif.json
If you find our work useful for your research, please cite our paper.
@article{yi2024diff,
title={Diff-IF: Multi-modality image fusion via diffusion model with fusion knowledge prior},
author={Yi, Xunpeng and Tang, Linfeng and Zhang, Hao and Xu, Han and Ma, Jiayi},
journal={Information Fusion},
pages={102450},
year={2024},
publisher={Elsevier}
}
If you have any questions, please send an email to [email protected].