Unsupervised Domain Adaptation for Semantic Segmentation of High-Resolution Remote Sensing Imagery Driven by Category-Certainty Attention

Pytorch implementation of our method for cross-domain semantic segmentation of the high-resolution remote sensing imagery.

Contact: Jingru Zhu (zhujingru@csu.edu.cn)

Paper

Unsupervised Domain Adaptation for Semantic Segmentation of High-Resolution Remote Sensing Imagery Driven by Category-Certainty Attention
Jie Chen , Member, IEEE, Jingru Zhu , Ya Guo , Geng Sun, Yi Zhang, and Min Deng
IEEE Transactions on Geoscience and Remote Sensing, 2022.

Please cite our paper if you find it useful for your research.

@inproceedings{UDAS_2022,
  author = {Jie Chen and Jingru Zhu and Ya Guo and Geng Sun and Yi Zhang and and Min Deng},
  booktitle = {IEEE Transactions on Geoscience and Remote Sensing},
  title = {Unsupervised Domain Adaptation for Semantic Segmentation of High-Resolution Remote Sensing Imagery Driven by Category-Certainty Attention},
  year = {2022}
}

Example Results

Quantitative Reuslts

Installation

• Install PyTorch from http://pytorch.org with Python 3.6 and PyTorch 1.8.0

• Clone this repo

git clone https://github.com/RS-CSU/UDAS-master
cd UDAS-master

Dataset

• Download the Potsdam Dataset as the source domain, and put it in the dataset/Potsdam folder

• Download the Vaihingen Dataset as the target domain, and put it in the data/Vaihingen folder

Testing

• Download the pre-trained Potsdam_best model and put it in the checkpoints_potsdam folder

• Download the pre-trained pot2vai model and put it in the checkpoints_pot2vai folder

• Test the model and results will be saved in the results folder

python test.py

Training Examples

• Train the Potsdam-to-Vaihingen model

python train_pot2vai_9_5.py

Related Implementation and Dataset

Acknowledgment

This code is heavily borrowed from Pytorch-AdaptSegNet.

Note

The model and code are available for non-commercial research purposes only.

• 07/2022: code released