SSL_CTT

This repository contains the official implementation of Semi-Supervised Semantic Segmentation with Cross Teacher Training, arxiv.

This code is based on ClassMix code

Abstract. Convolutional neural networks can achieve remarkable performance in semantic segmentation tasks. However, such neural network approaches heavily rely on costly pixel-level annotation. Semi-supervised learning is a promising resolution to tackle this issue, but its performance still far falls behind the fully supervised counterpart. This work proposes a cross-teacher training framework with three modules that significantly improves traditional semi-supervised learning approaches. The core is a cross-teacher module, which could simultaneously reduce the coupling among peer networks and the error accumulation between teacher and student networks. In addition, we propose two complementary contrastive learning modules. The high-level module can transfer high-quality knowledge from labeled data to unlabeled ones and promote separation between classes in feature space. The low-level module can encourage low-quality features learning from the high-quality features among peer networks. In experiments, the cross-teacher module significantly improves the performance of traditional student–teacher approaches, and our framework outperforms state-of-the-art methods on benchmark datasets.

Prerequisites

• CUDA/CUDNN
• Python3
• Packages found in requirements.txt

Datasets

mkdir ../dataset/CityScapes/

Download the dataset from here.

Pascal VOC 2012

mkdir ../dataset/VOC2012/

Download the dataset from here.

Experiments

For example, for Cityscapes with deeplabv2:

python3 cross_teacher_contr.py --config ./configs/configCityscapes.json --name deeplabv2_city

For example, for PASCAL VOC with deeplabv2:

python3 cross_teacher_contr.py --config ./configs/configVOC.json --name deeplabv2_voc

For example, for PASCAL VOC with deeplabv3+:

python3 cross_teacher_deeplabv3+.py --config ./configs/configCityscapes.json --name voc

Citation

@article{xiao2022semi,
  title={Semi-supervised semantic segmentation with cross teacher training},
  author={Xiao, Hui and Dong, Li and Xu, Hao and Fu, Shuibo and Yan, Diqun and Song, Kangkang and Peng, Chengbin},
  journal={Neurocomputing},
  volume={508},
  pages={36--46},
  year={2022},
  publisher={Elsevier}
}