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

Overcoming reliance on pixel-level heuristics such as confidence thresholding or entropy in test-time adaptation matters for anomaly segmentation because this paradigm has emerged as promising for mitigating distribution shifts in deep models.

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…