Multi-Level Label Correction by Distilling Proximate Patterns for Semi-supervised Semantic Segmentation

Abstract

Semi-supervised semantic segmentation relieves the reliance on large-scale labeled data by leveraging unlabeled data. Recent semi-supervised semantic segmentation approaches mainly resort to pseudo-labeling methods to exploit unlabeled data. However, unreliable pseudo-labeling can undermine the semi-supervision processes. In this paper, we propose an algorithm called Multi-Level Label Correction (MLLC), which aims to use graph neural networks to capture structural relationships in Semantic-Level Graphs (SLGs) and Class-Level Graphs (CLGs) to rectify erroneous pseudo-labels. Specifically, SLGs represent semantic affinities between pairs of pixel features, and CLGs describe classification consistencies between pairs of pixel labels. With the support of proximate pattern information from graphs, MLLC can rectify incorrectly predicted pseudo-labels and can facilitate discriminative feature representations. We design an end-to-end network to train and perform this effective label corrections mechanism. Experiments demonstrate that MLLC can significantly improve supervised baselines and outperforms state-of-the-art approaches in different scenarios on Cityscapes and PASCAL VOC 2012 datasets. Specifically, MLLC improves the supervised baseline by at least 5% and 2% with DeepLabV2 and DeepLabV3+ respectively under different partition protocols.

12 pages, 8 figures. IEEE Transactions on Multimedia, 2024

References in corpus (5)

• Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
• DMT: Dynamic Mutual Training for Semi-Supervised Learning
• Affinity Attention Graph Neural Network for Weakly Supervised Semantic Segmentation
• Conservative-Progressive Collaborative Learning for Semi-supervised Semantic Segmentation
• Plug-and-Play Pseudo Label Correction Network for Unsupervised Person Re-identification