Awesome

CorDA

Code for our paper Domain Adaptive Semantic Segmentation with Self-Supervised Depth Estimation ICCV21 <img src="./extra/fig.png" alt="alt text" width="50%" height="50%">

Prerequisite

Please create and activate the following conda envrionment. To reproduce our results, please kindly create and use this environment.

# It may take several minutes for conda to solve the environment
conda update conda
conda env create -f environment.yml
conda activate corda 

Code was tested on a V100 with 16G Memory.

Train a CorDA model

# Train for the SYNTHIA2Cityscapes task
bash run_synthia_stereo.sh
# Train for the GTA2Cityscapes task
bash run_gta.sh

Test a trained CorDA model

bash shells/eval_syn2city.sh 
bash shells/eval_gta2city.sh

Pre-trained models are provided (Google Drive). Please put them in ./checkpoint.

• The provided SYNTHIA2Cityscapes model achieves 56.3 mIoU (16 classes) at the end of the training.
• The provided GTA2Cityscapes model achieves 57.7 mIoU (19 classes) at the end of the training.

Reported Results on SYNTHIA2Cityscapes (The reported results are based on 5 runs instead of the best run.)

Citation

Please cite our work if you find it useful.

@inproceedings{wang2021domain,
  title={Domain Adaptive Semantic Segmentation with Self-Supervised Depth Estimation},
  author={Wang, Qin and Dai, Dengxin and Hoyer, Lukas and Van Gool, Luc and Fink, Olga},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  year={2021}
}

Acknowledgement

• DACS is used as our codebase and our DA baseline official
• SFSU as the source of stereo Cityscapes depth estimation Official

Data links

• Download links
  • Stereo Depth Estimation for Cityscapes
  • Mono Depth Estimation for GTA
  • SYNTHIA Depth and images SYNTHIA-RAND-CITYSCAPES (CVPR16)
• Dataset Folder Structure Tree

For questions regarding the code, please contact wang@qin.ee .