Exploring Object Relation in Mean Teacher for Cross-Domain Detection

This is the implementation of 'Exploring Object Relation in Mean Teacher for Cross-Domain Detection' [CVPR 2019]. The original paper can be found here.

Usage

1. Install mxnet. The version we use is 1.4.0.
2. Prepare the dataset. We mainly follow the steps in da-faster-rcnn.
3. Download the pre-trained models for Foggy Cityscapes and SIM10k. Then put them into models-foggy and models-sim10k.
4. Train Foggy Cityscapes domain adaptation or SIM-10k domain adaptation:

   ./train_foggy_final.sh  or ./train_sim10k_final.sh

5. The trained models for Foggy Cityscapes and SIM-10k are available at foggy_final (mAP=0.351)and sim10k_final (mAP=0.466).

Citation

If you find this code or model useful for your research, please cite our paper:

@inproceedings{cai2019exploring,
  title={Exploring Object Relation in Mean Teacher for Cross-Domain Detection},
  author={Cai, Qi and Pan, Yingwei and Ngo, Chong-Wah and Tian, Xinmei and Duan, Lingyu and Yao, Ting},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={11457--11466},
  year={2019}
}