DA-Cal: Towards Cross-Domain Calibration in Semantic Segmentation

2026-02-24

AI Summary

This work addresses the often-overlooked issue of model miscalibration in unsupervised domain adaptation (UDA) for semantic segmentation, where predicted confidence scores frequently misalign with actual accuracy—a critical concern in safety-sensitive applications. To tackle this, we propose DA-Cal, a novel framework that explicitly formulates cross-domain calibration as a soft pseudo-label optimization problem. DA-Cal introduces a meta temperature network to generate pixel-wise calibration parameters and integrates a bilevel optimization scheme with a complementary domain mixing strategy. Notably, our approach enhances both calibration quality and segmentation performance on the target domain without incurring additional inference overhead. Extensive experiments demonstrate significant improvements across multiple UDA semantic segmentation benchmarks, establishing DA-Cal as an effective solution for simultaneously improving accuracy and reliability in cross-domain settings.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

While existing unsupervised domain adaptation (UDA) methods greatly enhance target domain performance in semantic segmentation, they often neglect network calibration quality, resulting in misalignment between prediction confidence and actual accuracy -- a significant risk in safety-critical applications. Our key insight emerges from observing that performance degrades substantially when soft pseudo-labels replace hard pseudo-labels in cross-domain scenarios due to poor calibration, despite the theoretical equivalence of perfectly calibrated soft pseudo-labels to hard pseudo-labels. Based on this finding, we propose DA-Cal, a dedicated cross-domain calibration framework that transforms target domain calibration into soft pseudo-label optimization. DA-Cal introduces a Meta Temperature Network to generate pixel-level calibration parameters and employs bi-level optimization to establish the relationship between soft pseudo-labels and UDA supervision, while utilizing complementary domain-mixing strategies to prevent overfitting and reduce domain discrepancies. Experiments demonstrate that DA-Cal seamlessly integrates with existing self-training frameworks across multiple UDA segmentation benchmarks, significantly improving target domain calibration while delivering performance gains without inference overhead. The code will be released.

Problem

• unsupervised domain adaptation
• semantic segmentation
• model calibration
• cross-domain calibration
• pseudo-labels

Innovation

• cross-domain calibration
• semantic segmentation
• unsupervised domain adaptation
• soft pseudo-labels
• meta temperature network