DMT: Dynamic mutual training for semi-supervised learning

作者：

Highlights：

• A new viewpoint from inter-model disagreement to alleviate pseudo label noise in semi-supervised learning.

• Quantify inter-model disagreement and propose dynamic mutual training with a noise-robust loss.

• State-of-the-art performance in semi-supervised image classification and semantic segmentation.

摘要

•A new viewpoint from inter-model disagreement to alleviate pseudo label noise in semi-supervised learning.•Quantify inter-model disagreement and propose dynamic mutual training with a noise-robust loss.•State-of-the-art performance in semi-supervised image classification and semantic segmentation.

论文关键词：Dynamic mutual training,Inter-model disagreement,Noisy pseudo label,Semi-supervised learning

论文评审过程：Received 16 January 2021, Revised 19 August 2021, Accepted 26 November 2021, Available online 11 May 2022, Version of Record 9 June 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108777