/LCGDM

[TGRS 2023] Official implementation for the paper Domain Adaptive Land-Cover Classification via Local Consistency and Global Diversity.

Primary LanguagePython

Domain Adaptive Land-Cover Classification via Local Consistency and Global Diversity

This is the official implementation for the TGRS 2023 paper Domain Adaptive Land-Cover Classification via Local Consistency and Global Diversity.

Dependencies

Training with different UDA methods (baseline, +Ent, +LCGDM)

For example, to reproduce the results of CBST in the paper, one can run

bash scripts/cbst/train_cbst.sh #baseline
bash scripts/cbst/train_cbst_Ent.sh # +Ent
bash scripts/cbst/train_cbst_SL.sh # +LCGDM

Inference on the test set

Submit your test results on LoveDA Unsupervised Domain Adaptation Challenge and obtain the final score.

python predict.py # you should set the arguments (e.g. model path)

T-SNE visualization

python tsne.py # CBST and CLAN are supported, you should set the model path and the image path

Hyper-parameters Configuration

Detailed hyperparameters config can be found in folder "configs/LoveDA".

Citation

If you use our code in your research, please cite our TGRS 2023 paper.

@article{DBLP:journals/tgrs/MaZWZ23,
  author       = {Ailong Ma and
                  Chenyu Zheng and
                  Junjue Wang and
                  Yanfei Zhong},
  title        = {Domain Adaptive Land-Cover Classification via Local Consistency and
                  Global Diversity},
  journal      = {{IEEE} Trans. Geosci. Remote. Sens.},
  volume       = {61},
  pages        = {1--17},
  year         = {2023}
}

Acknowledgments

The code is developed based on the following repositories. We appreciate their nice implementations.

Method Repository
LoveDA https://github.com/Junjue-Wang/LoveDA
LoveCS https://github.com/Junjue-Wang/LoveCS
DCA https://github.com/Luffy03/DCA