Learning pseudo labels for semi-and-weakly supervised semantic segmentation

作者：

Highlights：

• We improve the semi-and-weakly supervised semantic segmentation via learning high-quality pseudo labels.

• A simpler learning target for pseudo label generation is introduced to void overfitting.

• The interaction between two networks progressively produces additional self-supervision to improve representation learning.

• Our method outperforms the state-of-the-art methods significantly.

摘要

•We improve the semi-and-weakly supervised semantic segmentation via learning high-quality pseudo labels.•A simpler learning target for pseudo label generation is introduced to void overfitting.•The interaction between two networks progressively produces additional self-supervision to improve representation learning.•Our method outperforms the state-of-the-art methods significantly.

论文关键词：Semi-supervised,Weakly supervised,Semi-and-weakly supervised,Semantic segmentation,Pseudo label,Self-training

论文评审过程：Received 17 October 2021, Revised 6 May 2022, Accepted 21 July 2022, Available online 23 July 2022, Version of Record 2 August 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108925