Skeleton-Guided Progressive Test-Time Adaptation for Thin Curvilinear Structures

Boa Jang; JunGyu Lee; Gwanho Lee; Jinwook Choi; Young-Gon Kim

cs.CVOct 8, 2026

Skeleton-Guided Progressive Test-Time Adaptation for Thin Curvilinear Structures

Authors: Boa Jang, JunGyu Lee, Gwanho Lee, Jinwook Choi, Young-Gon Kim

Organizations: Seoul National University · Seoul National University Hospital · Yonsei University · Korea University

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.

Figures & tables

Figure 1: Representative curvilinear structure datasets and their domain characteristics. (a) Images from color fundus photography (CFP), OCT angiography (OCTA), fluorescein angiography (FA), and road datasets. (b) t-SNE of batch-normalization statistics; colors denote datasets, marker shapes denote imaging domains.

Figure 2: Overview of SGP-TTA. A source-trained network with frozen convolutions adapts only at its BN layers, through two complementary mechanisms. ProgBN blends frozen source and current target statistics under a sample-count schedule, without gradient updates, while CSR recalls the skeleton of the multi-view consensus map and updates only the BN affine parameters.

Method	DRIVE $\rightarrow$ STARE		DRIVE $\rightarrow$ CHASEDB		OCTA3mm $\rightarrow$ ROSE1		OCTA6mm $\rightarrow$ ROSE1		ROSE1 $\rightarrow$ OCTA3mm		ROSE1 $\rightarrow$ OCTA6mm
	Dice	clDice	Dice	clDice	Dice	clDice	Dice	clDice	Dice	clDice	Dice	clDice
Source	51.32	48.55	37.08	38.95	47.47	41.75	56.49	49.72	59.76	64.22	68.92	80.51
TENT ( Wang et al. 2020 )	70.96	68.66	71.60	73.51	48.47	41.65	53.77	48.79	63.17	66.07	71.71	80.89
CoTTA ( Wang et al. 2022b )	71.67	69.66	71.27	73.43	47.54	42.75	53.94	48.95	62.26	64.52	71.41	80.30
SAR ( Niu et al. 2023 )	70.96	68.69	71.37	73.22	48.55	41.73	53.84	48.87	62.17	64.44	71.36	80.26
EATA ( Niu et al. 2022 )	71.06	68.78	71.61	73.56	48.57	43.73	53.87	49.79	62.44	65.05	71.42	80.60

Table 1: Comparison with TTA methods for curvilinear structure segmentation under within-modality shifts across retinal imaging datasets. Each column denotes a transfer setting. Source denotes the source model tested on the target domain without adaptation. The best and second-best results in each column are highlighted in bold and underline , respectively.

Table 4

Figure 3: Qualitative comparison across diverse target domains. Red boxes indicate restored continuity of fine vessel structures and reduced false negatives in our SGP-TTA.

Table 6

Figure 4: Effect of the $\alpha$ schedule over the target stream. Improvement over the source model against the number of observed target samples $N$ , averaged over all settings.

Figure 5: Robustness across segmentation architectures. Mean clDice on five different BN-based backbones, averaged over all settings.

Figure 6: clDice against adaptation cost. Mean inference time per target image and clDice, averaged over all settings; marker area scales with the number of updated parameters.

Explore similar work

Learning Topology-Aware Representations via Test-Time Adaptation for Anomaly Segmentation
Jun 26, 2026Ali Zia, Usman Ali, Abdul Rehman +5Test-Time Adaptation Unsupervised Anomaly Detection
School of Computing, Engineering and Mathematical Sciences, La Trobe University, Melbourne, Australia · School of Engineering and Applied Sciences, GIFT University, Gujranwala 52250, Pakistan · Sino-Pak Centre for Artificial Intelligence Pak-Austria Fachhochschule Institute of Applied Sciences and Technology
TestMate: Test-Time Domain Adaptation Aided by Lightweight Vision Foundation Model
Jul 4, 2026Dimitrios Fotiou, Vasileios Mygdalis, Ioannis PitasSource-Free Domain Adaptation Test-Time Adaptation
Aristotle University of Thessaloniki
Memory-Supported Synergistic Adaptation for Training-Free Test-Time Medical Image Segmentation
Jul 20, 2026Lingrui Li, Nan Pu, Dong Zhao +4Memory-Augmented VLMs VLM Adaptation
School of Computer Science, University of Nottingham, UK · School of Computer Science and Information Engineering, Hefei University of Technology, China · Information Systems Technology and Design Pillar, Singapore University of Technology and Design, Singapore

Jun 26, 2026, cs.CVJ/K move · Enter open · S save

Learning Topology-Aware Representations via Test-Time Adaptation for Anomaly Segmentation

Ali Zia, Usman Ali, Abdul Rehman, Umer Ramzan+4

School of Computing, Engineering and Mathematical Sciences, La Trobe University, Melbourne, Australia · School of Engineering and Applied Sciences, GIFT University, Gujranwala 52250, Pakistan · Sino-Pak Centre for Artificial Intelligence Pak-Austria Fachhochschule Institute of Applied Sciences and Technology

Test-Time Adaptation Unsupervised Anomaly Detection Anomaly Detection

Open paper AskPDF arXiv

Test-time adaptation (TTA) has emerged as a promising paradigm for mitigating distribution shifts in deep models. However, existing TTA approaches for anomaly segmentation remain limited by their reliance on pixel-level heuristics, such as confidence thresholding or entropy minimisation, which fail to preserve structural consistency under noise and texture variation. Moreover, they typically treat anomaly maps as flat intensity fields, ignoring the higher-order spatial relationships that characterise complex defect geometries. We introduce TopoTTA (Topological Test-Time Adaptation), a novel framework that integrates persistent homology, a tool from topological data analysis, into the TTA pipeline to enforce geometric and structural coherence during adaptation. By applying multi-level cubical complex filtration to anomaly score maps, TopoTTA derives robust topological pseudo-labels that guide a lightweight test-time classifier, enhancing segmentation quality without retraining the backbone model. The approach avoids reliance on method-specific raw-score thresholding for mask binarisation, preserves connectivity, and generalises across both 2D and 3D modalities. Extensive experiments across six standard benchmarks (MVTec AD, VisA, Real-IAD, MVTec 3D-AD, AnomalyShapeNet, and MVTec LOCO) demonstrate an average 15% F1 improvement over state-of-the-art unsupervised anomaly detection and segmentation methods, with the largest gains on anomalies exhibiting complex geometric or structural variations. These findings suggest that integrating topological reasoning into test-time adaptation provides a principled route to structure-aware generalisation, bridging the gap between geometric learning and robust adaptation.

Abstract

Explore similar work

Learning Topology-Aware Representations via Test-Time Adaptation for Anomaly Segmentation

TestMate: Test-Time Domain Adaptation Aided by Lightweight Vision Foundation Model

Memory-Supported Synergistic Adaptation for Training-Free Test-Time Medical Image Segmentation