/ADS-SemiSeg

Adversarial Dual-Student with Differentiable Spatial Warping for Semi-Supervised Semantic Segmentation. TCSVT 2022

Primary LanguagePython

ADS-SemiSeg

This repository contains official implementation of Adversarial Dual-Student with Differentiable Spatial Warping for Semi-Supervised Semantic Segmentation in TCSVT 2022, by Cong Cao, Tianwei Lin, Dongliang He, Fu Li, Huanjing Yue, Jingyu Yang, and Errui Ding. [arxiv] [journal]

Code

Environment

  • Python >= 3.5
  • Pytorch >= 1.1
  • NVIDIA Tesla V100

Test

You can download pretrained weights from here (ADS-DGW_Dataset_SemiRatio_iterXXXXX.pth), then run:

bash run_scripts/test_VOC2012.sh

Train

Train baseline:

bash run_scripts/train_baseline_VOC2012.sh

Train Mean-Teacher with DGW augmentation:

bash run_scripts/train_MT_DGW_VOC2012.sh

Train ADS with DGW augmentation:

bash run_scripts/train_ADS_DGW_VOC2012.sh

Citation

If you find our paper or code helpful in your research or work, please cite our paper:

@article{cao2022adversarial,
  title={Adversarial dual-student with differentiable spatial warping for semi-supervised semantic segmentation},
  author={Cao, Cong and Lin, Tianwei and He, Dongliang and Li, Fu and Yue, Huanjing and Yang, Jingyu and Ding, Errui},
  journal={IEEE Transactions on Circuits and Systems for Video Technology},
  volume={33},
  number={2},
  pages={793--803},
  year={2022},
  publisher={IEEE}
}