English

DMT: Dynamic Mutual Training for Semi-Supervised Learning

Computer Vision and Pattern Recognition 2022-05-12 v4

Abstract

Recent semi-supervised learning methods use pseudo supervision as core idea, especially self-training methods that generate pseudo labels. However, pseudo labels are unreliable. Self-training methods usually rely on single model prediction confidence to filter low-confidence pseudo labels, thus remaining high-confidence errors and wasting many low-confidence correct labels. In this paper, we point out it is difficult for a model to counter its own errors. Instead, leveraging inter-model disagreement between different models is a key to locate pseudo label errors. With this new viewpoint, we propose mutual training between two different models by a dynamically re-weighted loss function, called Dynamic Mutual Training (DMT). We quantify inter-model disagreement by comparing predictions from two different models to dynamically re-weight loss in training, where a larger disagreement indicates a possible error and corresponds to a lower loss value. Extensive experiments show that DMT achieves state-of-the-art performance in both image classification and semantic segmentation. Our codes are released at https://github.com/voldemortX/DST-CBC .

Keywords

Cite

@article{arxiv.2004.08514,
  title  = {DMT: Dynamic Mutual Training for Semi-Supervised Learning},
  author = {Zhengyang Feng and Qianyu Zhou and Qiqi Gu and Xin Tan and Guangliang Cheng and Xuequan Lu and Jianping Shi and Lizhuang Ma},
  journal= {arXiv preprint arXiv:2004.08514},
  year   = {2022}
}

Comments

Published at Pattern Recognition, see https://www.sciencedirect.com/science/article/abs/pii/S0031320322002588

R2 v1 2026-06-23T14:55:58.612Z