PanBench: Towards High-Resolution and High-Performance Pansharpening

Introduction

This repository is the official PyTorch implementation of our paper PanBench: Towards High-Resolution and High-Performance Pansharpening.

Requirements

To install dependencies:

pip install -r requirements.txt

Dataset

Google Drive | 123 Cloud Drive

PanBench
├─GF1
│  ├─NIR_256
│  ├─PAN_1024
│  └─RGB_256
├─GF2
│  ├─NIR_256
│  ├─PAN_1024
│  └─RGB_256
├─GF6
│  ├─NIR_256
│  ├─PAN_1024
│  └─RGB_256
├─IN
│  ├─NIR_256
│  ├─PAN_1024
│  └─RGB_256
├─LC7
│  ├─NIR_256
│  ├─PAN_1024
│  └─RGB_256
├─LC8
│  ├─NIR_256
│  ├─PAN_1024
│  └─RGB_256
├─QB
│  ├─NIR_256
│  ├─PAN_1024
│  └─RGB_256
├─WV2
│  ├─NIR_256
│  ├─PAN_1024
│  └─RGB_256
├─WV3
│  ├─NIR_256
│  ├─PAN_1024
│  └─RGB_256
└─WV4
    ├─NIR_256
    ├─PAN_1024
    └─RGB_256

Training

To train the models in the paper, run these commands:

python src/train.py experiment=cmfnet

Testing

To test the models in the paper, run these commands:

python src/train.py experiment=cmfnet

Evaluation

To evaluate the models in the paper, run these commands:

python src/eval.py experiment=cmfnet

Pre-trained Models

You can download pretrained models here: CMFNet.

Overview of Model Zoo and Datasets

We support various pansharpening methods and satellites. We are working on add new methods and collecting experiment results.

Currently supported methods.
- PNN (Remote Sensing'2016)
- PanNet (ICCV'2017)
- MSDCNN (J-STARS'2018)
- TFNet (Inform. Fusion'2020)
- FusionNet (TGRS'2020)
- GPPNN (CVPR'2021)
- SRPPNN (TGRS'2021)
- PGCU (CVPR'2023)
Currently supported satellites.
- GaoFen1
- GaoFen2
- GaoFen6
- Landsat7
- Landsat8
- WorldView2
- WorldView3
- WorldView4
- QuickBird
- IKONOS

Visualization

We provide a visualization tool to help you understand the training process. You can use the following command to start the visualization tool.

python visualize.py

Citation

If you use our code or models in your research, please cite with:

@misc{wang2023panbench,
      title={PanBench: Towards High-Resolution and High-Performance Pansharpening}, 
      author={Shiying Wang and Xuechao Zou and Kai Li and Junliang Xing and Pin Tao},
      year={2023},
      eprint={2311.12083},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

XavierJiezou/Pansharpening