Affinity Space Adaptation for Semantic Segmentation Across Domains.

Area of Science:

• Computer Vision
• Deep Learning
• Machine Learning

Background:

• Deep learning excels in semantic segmentation with dense annotations.
• Generalizing semantic segmentation to diverse, real-world scenarios remains a significant challenge.
• Unsupervised Domain Adaptation (UDA) aims to bridge the gap between labeled source domains and unlabeled target domains.

Purpose of the Study:

• To develop a novel approach for unsupervised domain adaptation in semantic segmentation.
• To leverage invariant semantic structures across domains by analyzing pairwise pixel co-occurrence patterns.
• To improve the generalization capability of semantic segmentation models in the wild.

Main Methods:

• Proposing a method that adapts domains based on the affinity relationship between adjacent pixels (affinity space).
• Introducing two key strategies: affinity space cleaning and adversarial affinity space alignment.
• Moving beyond traditional pixel-wise adaptation by focusing on structured relationships.

Main Results:

• The proposed method demonstrates superior performance compared to state-of-the-art approaches on challenging cross-domain semantic segmentation benchmarks.
• The approach effectively exploits invariant semantic structures for improved domain adaptation.
• Validation through extensive experiments on multiple datasets.

Conclusions:

• The affinity space adaptation method offers a promising direction for unsupervised domain adaptation in semantic segmentation.
• The technique enhances model robustness and generalization across different domains.
• The developed strategies provide effective solutions for adapting models without target domain annotations.