LGCT: Local–Global Collaborative Transformer for Fusion of Hyperspectral and Multispectral Images (TGRS 2024)

Official code repo for LGCT: Local–Global Collaborative Transformer for Fusion of Hyperspectral and Multispectral Images (TGRS'24) [PDF]

By Wangquan He, Xiyou Fu, Nanying Li, Qi Ren and Sen Jia.

✨ Updates

• 2025.08.12: We have released a version that removes the img_size parameter (/models/LGCT_arch_×4/8_ar_v1.py). Now supports images of any size automatically.
• 2025.07.22: We have uploaded the ×8 version of LGCT model (/models/LGCT_arch_×8.py) for fusion tasks.

Network

Requirements

To install dependencies:

# create new Anaconda env
conda create -n LGCT python=3.8 -y
conda activate LGCT

# install python dependencies
pip install -r requirements.txt

Usage:

Before training, you need to:

• Download Datasets : Pavia University / Houston / Chikusei.

• Set the data path -root in args_parser.py.

1. Training

Use the following command to train the network. Take training LGCT ×4 in the Houston dataset as an example,

 CUDA_VISIBLE_DEVICES=0 python -u main.py  \
    -arch 'LGCT' \
    -root '[root path of datasets]' \
    -dataset 'Houston' \
    --scale_ratio 4 \
    --model_path './checkpoints'\
    --n_epochs 10000 --lr 1e-4\
    --criterion 'L1' \

2. Testing

Before testing, set the pre-trained model weight 'pth' files to line 57 in test.py. These files are obtained through the training phase and can be found in the ./checkpoints.

Then run the following command:

python test.py

Citation

If you find this work helpful, please consider citing it. We would greatly appreciate it!

@article{he2024lgct,
  title={LGCT: Local-Global Collaborative Transformer for Fusion of Hyperspectral and Multispectral Images},
  author={He, Wangquan and Fu, Xiyou and Li, Nanying and Ren, Qi and Jia, Sen},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  year={2024},
  volume={62},
  number={},
  pages={1-14},
  publisher={IEEE}
}

Others

For any questions, please contact us (hewangquan2022@email.szu.edu.cn).