Scale Aware Adaptation for Land-Cover Classificationin Remote Sensing Imagery (in PyTorch)

This repository provides the ResNet-101-based segmentation model trained on ISPRS dataset from the paper Scale Aware Adaptation for Land-Cover Classificationin Remote Sensing Imagery (the provided weights achieve 47.66% mean IoU on the validation set for Potsdam to Vaihingen)

Train the model

$ python train_scaleDA_isprs.py --folder DATA-ROOT

Test the model

$ python test_isprs.py --resume CHECKPOINT 

Please cite our paper:

@inproceedings{deng2021scale,
  title={Scale Aware Adaptation for Land-Cover Classification in Remote Sensing Imagery},
  author={Deng, Xueqing and Zhu, Yi and Tian, Yuxin and Newsam, Shawn},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={2160--2169},
  year={2021}
}