Skeleton-guided method improves segmentation of thin structures under domain shifts
Skeleton-Guided Progressive Test-Time Adaptation for Thin Curvilinear Structures
Computer Vision and Pattern Recognition
Summary
Segmenting thin lines like blood vessels or roads in images is tough because even tiny mistakes can break their shape. When images come from different devices or settings, these mistakes get worse. The authors created a method that uses the underlying shape, or skeleton, of these lines to guide the adjustment process during testing, making the results more connected and accurate. Their approach changes how the model normalizes data step-by-step and uses multiple views to ensure the main structure stays intact.
What this means in practice
- •For medical imaging teams: Improve automatic vessel segmentation accuracy in images from different scanning devices by preserving vessel connectivity during model adaptation without source data.
- •For geospatial mapping teams: Enhance road extraction from satellite imagery across different sensor types by using skeleton-based adaptation to maintain continuous road networks.
Authors
Boa Jang, JunGyu Lee, Gwanho Lee, Jinwook Choi, Young-Gon Kim
Abstract
Accurate segmentation of thin curvilinear structures is vital for various real-world applications, from vessel analysis to road extraction. Yet their intricate geometry makes even minor pixel-wise errors enough to break the global topology, and this structural fragility turns severe domain shifts into catastrophic failures. The difficulty is most acute under cross-modality gaps, where the imaging process itself differs fundamentally between source and target. While test-time adaptation (TTA) offers a practical source-free remedy, existing methods adapt feature statistics and confidence, neither of which constrains connectivity, and thus degrade under such extreme gaps. To address this, we propose Skeleton-Guided Progressive Test-Time Adaptation (SGP-TTA). Progressive Batch Normalization (ProgBN) shifts normalization from frozen source statistics toward current target estimates under a sample-count schedule, so that the source-target balance follows the stage of adaptation rather than a fixed coefficient. Consensus Skeleton Recall (CSR) then derives a structural target from geometrically aligned multi-view predictions and updates only the BN affine parameters to preserve connected structures. Extensive experiments show that SGP-TTA consistently outperforms existing TTA methods in topological connectivity, with the largest margins under cross-modality shift. The project page is available at https://boa-jang.github.io/SGP-TTA.