DCLDA

IEEE JSTAR 2024 code

This repository is the implementation of the following two papers:

1. Domain Adaptation with Contrastive Learning for Object Detection in Satellite Imagery
2. Unsupervised Domain Adaptation With Debiased Contrastive Learning and Support-Set Guided Pseudolabeling for Remote Sensing Images

Installation

1. Create and activate a new conda environment (Recommended to use Python>=3.9)
2. Install PyTorch & cuda from official website
   Note: It is recommended to install CUDA version >= 11.0
3. Clone the repository in local

$ cd DCLDA
$ python -m pip install -e CenterNet2
$ cd CenterNet2
$ pip install -r requirements.txt

Architecture

Train

train_net.py allows you to train new model from strach.

$ cd /home/username/aerialAdaptation/CenterNet2/projects/CenterNet2
$ python train_net_multi_datasets_Satellite.py for DIOR to DOTA dataset UDA training
$ python train_net_multi_datasets_UAV.py for Vsidrone to UAVDT dataset UDA training 

Validation

validation.py allows you to run inference on trained a model. User need to provide the annotations(COCO json format) and images

$ cd /home/username/aerialAdaptation/CenterNet2/projects/CenterNet2
$ python validation.py for all dataset inference

Experimental Results

The overall and classwise performance considering DIOR as the source and DOTA as the target dataset is provided below:

The overall and classwise performance considering VISDRONE as the source and UAVDT as the target dataset is provided below:

License

Our code repository is under Apache 2.0 license. Codes Under projects/CenterNet2 are also under Apache 2.0 license. projects/CenterNet2/centernet/modeling/backbone/bifpn_fcos.py are from AdelaiDet, which follows the original non-commercial license. The code from detectron2 follows the original Apache 2.0 license.

Citation

If you find this project useful for your research, please use the following BibTeX entry.

@article{biswas2024unsupervised,
  title={Unsupervised domain adaptation with debiased contrastive learning and support-set guided pseudo labeling for remote sensing images},
  author={Biswas, Debojyoti and Te{\v{s}}i{\'c}, Jelena},
  journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
  year={2024},
  publisher={IEEE}
}

@inproceedings{zhou2021probablistic,
  title={Probabilistic two-stage detection},
  author={Zhou, Xingyi and Koltun, Vladlen and Kr{\"a}henb{\"u}hl, Philipp},
  booktitle={arXiv preprint arXiv:2103.07461},
  year={2021}
}