PMAA: A Progressive Multi-scale Attention Autoencoder Model for High-Performance Cloud Removal from Multi-temporal Satellite Imagery
This repository is the official PyTorch implementation of the accepted paper PMAA of ECAI 2023 Oral.
Xuechao Zou1,*, Kai Li2,*, Junliang Xing2, Pin Tao1,2,†, Yachao Cui1
Qinghai University1 • Tsinghua University2
- [2023/07/30] Code release.
- [2023/07/16] PMAA got accepted by ECAI 2023 Oral.
- [2023/03/29] PMAA is on arXiv now.
To install dependencies:
pip install -r requirements.txt
To download datasets:
-
Sen2_MTC_Old: multipleImage.tar.gz
-
Sen2_MTC_New: CTGAN.zip
To train the models in the paper, run these commands:
python train_old.py
python train_new.py
To evaluate my models on two datasets, run:
python test_old.py
python test_new.py
You can download pretrained models here:
- Our awesome model trained on Sen2_MTC_old: pmaa_old.pth
- Our awesome model trained on Sen2_MTC_new: pmaa_new.pth
If you use our code or models in your research, please cite with:
@article{zou2023pmaa,
title={PMAA: A Progressive Multi-scale Attention Autoencoder Model for High-Performance Cloud Removal from Multi-temporal Satellite Imagery},
author={Zou, Xuechao and Li, Kai and Xing, Junliang and Tao, Pin and Cui, Yachao},
journal={European Conference on Artificial Intelligence (ECAI)},
year={2023},
pages={3165-3172},
}