Official implementation for
Learning Remote Sensing Aleatoric Uncertainty for Semi-Supervised Change Detection
- LEVIR-AU: this dataset is based on LEVIR, and it is released in GoogleDrive: levir-au
- Pretrained model: backbone-resnet50, and put this file under the dir: ./pretrained/
pip install -r requirements.txt{YOUR DATA ROOT}
├── A
│ ├── x.jpg
│ └── xx.jpg
│
├── B
│ ├── x.jpg
│ └── xx.jpg
│
└── label
├── x.jpg
└── xx.jpg
-----------------------------------------------
# train/test files
corresponding split txt at: {YOUR PROJECT ROOT}/splits
For example, use the LEVIR-AU dataset, then modify:
{YOUR PROJECT ROOT}/configs/levir_au.yaml
CUDA_VISIBLE_DEVICES=1 sh scripts/train_semi.sh 1 25000To refer to this work, please cite
@ARTICLE{10621657,
author={Shen, Jinhao and Zhang, Cong and Zhang, Mingwei and Li, Qiang and Wang, Qi},
journal={IEEE Transactions on Geoscience and Remote Sensing},
title={Learning Remote Sensing Aleatoric Uncertainty for Semi-Supervised Change Detection},
year={2024},
volume={62},
number={},
pages={1-13},
keywords={Uncertainty;Remote sensing;Unified modeling language;Training;Imaging;Task analysis;Pipelines;Change detection (CD);remote sensing;semi-supervised learning;uncertainty},
doi={10.1109/TGRS.2024.3437250}}
Thanks for the open source codes: SemiCD, FPA, Unimatch, RCL.