Semi-Supervised Semantic Segmentation via Marginal Contextual Information

TL;DR

This paper introduces S4MC, a semi-supervised semantic segmentation method that improves pseudo label quality by leveraging spatial correlations among neighboring pixels, leading to better performance with minimal additional computation.

Contribution

The paper proposes a novel confidence refinement scheme that uses marginal contextual information to enhance pseudo labels in semi-supervised segmentation, outperforming existing methods.

Findings

S4MC achieves a 1.39 mIoU improvement on PASCAL VOC 12.

The method maintains high pseudo label quality with negligible computational overhead.

S4MC outperforms state-of-the-art semi-supervised segmentation approaches.

Abstract

We present a novel confidence refinement scheme that enhances pseudo labels in semi-supervised semantic segmentation. Unlike existing methods, which filter pixels with low-confidence predictions in isolation, our approach leverages the spatial correlation of labels in segmentation maps by grouping neighboring pixels and considering their pseudo labels collectively. With this contextual information, our method, named S4MC, increases the amount of unlabeled data used during training while maintaining the quality of the pseudo labels, all with negligible computational overhead. Through extensive experiments on standard benchmarks, we demonstrate that S4MC outperforms existing state-of-the-art semi-supervised learning approaches, offering a promising solution for reducing the cost of acquiring dense annotations. For example, S4MC achieves a 1.39 mIoU improvement over the prior art on PASCAL VOC 12 with 366 annotated images. The code to reproduce our experiments is available at https://s4mcontext.github.io/