Domain Generalization for Semantic Segmentation: A Survey

2025-10-03

AI Summary

Domain Generalization (DG) for semantic segmentation aims to enhance model generalization to unseen target domains—critical for safety-critical applications such as autonomous driving and biomedical imaging. This paper presents a systematic survey of DG semantic segmentation methods and introduces the first unified taxonomy tailored to this task, explicitly characterizing the paradigm shift from conventional approaches—including domain-invariant feature learning, data augmentation, and meta-learning—to fine-tuning vision foundation models (VFMs). Through extensive empirical evaluation across diverse domain-shift benchmarks, we demonstrate that VFM-based methods consistently achieve superior cross-domain generalization, particularly under large domain discrepancies. Our analysis clarifies the technical evolution of DG semantic segmentation and establishes VFMs as a pivotal enabler for next-generation solutions. This work not only provides a structured conceptual framework but also catalyzes the transition of DG semantic segmentation toward foundation-model-driven methodologies.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Deep Neural Architectures and Foundation Models
• Application Domains: Security

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: Foundation models and LLMs for Web-related graphs

Abstract

The generalization of deep neural networks to unknown domains is a major challenge despite their tremendous progress in recent years. For this reason, the dynamic area of domain generalization (DG) has emerged. In contrast to unsupervised domain adaptation, there is no access to or knowledge about the target domains, and DG methods aim to generalize across multiple different unseen target domains. Domain generalization is particularly relevant for the task semantic segmentation which is used in several areas such as biomedicine or automated driving. This survey provides a comprehensive overview of the rapidly evolving topic of domain generalized semantic segmentation. We cluster and review existing approaches and identify the paradigm shift towards foundation-model-based domain generalization. Finally, we provide an extensive performance comparison of all approaches, which highlights the significant influence of foundation models on domain generalization. This survey seeks to advance domain generalization research and inspire scientists to explore new research directions.

Problem

• Surveying domain generalization methods for semantic segmentation tasks
• Addressing generalization to unseen domains without target data access
• Analyzing foundation models' impact on cross-domain generalization performance

Innovation

• Surveying domain generalization semantic segmentation methods
• Identifying shift towards foundation-model-based approaches
• Comparing performance highlighting foundation models' influence