Rethinking Semi-supervised Segmentation Beyond Accuracy: Reliability and Robustness

TL;DR

This paper highlights the importance of evaluating semi-supervised segmentation models not only on accuracy but also on reliability and robustness, proposing a new comprehensive metric called RSS for better real-world applicability.

Contribution

It introduces the Reliable Segmentation Score (RSS), a novel metric that combines accuracy, calibration, and uncertainty to holistically evaluate segmentation models.

Findings

Semi-supervised models often sacrifice reliability for accuracy.

UniMatchV2 shows robustness but still has reliability issues.

Holistic evaluation metrics like RSS are essential for real-world deployment.

Abstract

Semantic segmentation is critical for scene understanding but demands costly pixel-wise annotations, attracting increasing attention to semi-supervised approaches to leverage abundant unlabeled data. While semi-supervised segmentation is often promoted as a path toward scalable, real-world deployment, it is astonishing that current evaluation protocols exclusively focus on segmentation accuracy, entirely overlooking reliability and robustness. These qualities, which ensure consistent performance under diverse conditions (robustness) and well-calibrated model confidences as well as meaningful uncertainties (reliability), are essential for safety-critical applications like autonomous driving, where models must handle unpredictable environments and avoid sudden failures at all costs. To address this gap, we introduce the Reliable Segmentation Score (RSS), a novel metric that combines predictive accuracy, calibration, and uncertainty quality measures via a harmonic mean. RSS penalizes deficiencies in any of its components, providing an easy and intuitive way of holistically judging segmentation models. Comprehensive evaluations of UniMatchV2 against its predecessor and a supervised baseline show that semi-supervised methods often trade reliability for accuracy. While out-of-domain evaluations demonstrate UniMatchV2's robustness, they further expose persistent reliability shortcomings. We advocate for a shift in evaluation protocols toward more holistic metrics like RSS to better align semi-supervised learning research with real-world deployment needs.