Instance-Aware Test-Time Segmentation for Continual Domain Shifts

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lee, Seunghwan, Jung, Inyoung, Lee, Hojoon, Park, Eunil, Hong, Sungeun
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909951807455232
author Lee, Seunghwan
Jung, Inyoung
Lee, Hojoon
Park, Eunil
Hong, Sungeun
author_facet Lee, Seunghwan
Jung, Inyoung
Lee, Hojoon
Park, Eunil
Hong, Sungeun
contents Continual Test-Time Adaptation (CTTA) enables pre-trained models to adapt to continuously evolving domains. Existing methods have improved robustness but typically rely on fixed or batch-level thresholds, which cannot account for varying difficulty across classes and instances. This limitation is especially problematic in semantic segmentation, where each image requires dense, multi-class predictions. We propose an approach that adaptively adjusts pseudo labels to reflect the confidence distribution within each image and dynamically balances learning toward classes most affected by domain shifts. This fine-grained, class- and instance-aware adaptation produces more reliable supervision and mitigates error accumulation throughout continual adaptation. Extensive experiments across eight CTTA and TTA scenarios, including synthetic-to-real and long-term shifts, show that our method consistently outperforms state-of-the-art techniques, setting a new standard for semantic segmentation under evolving conditions.
format Preprint
id arxiv_https___arxiv_org_abs_2512_08569
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Instance-Aware Test-Time Segmentation for Continual Domain Shifts
Lee, Seunghwan
Jung, Inyoung
Lee, Hojoon
Park, Eunil
Hong, Sungeun
Computer Vision and Pattern Recognition
Continual Test-Time Adaptation (CTTA) enables pre-trained models to adapt to continuously evolving domains. Existing methods have improved robustness but typically rely on fixed or batch-level thresholds, which cannot account for varying difficulty across classes and instances. This limitation is especially problematic in semantic segmentation, where each image requires dense, multi-class predictions. We propose an approach that adaptively adjusts pseudo labels to reflect the confidence distribution within each image and dynamically balances learning toward classes most affected by domain shifts. This fine-grained, class- and instance-aware adaptation produces more reliable supervision and mitigates error accumulation throughout continual adaptation. Extensive experiments across eight CTTA and TTA scenarios, including synthetic-to-real and long-term shifts, show that our method consistently outperforms state-of-the-art techniques, setting a new standard for semantic segmentation under evolving conditions.
title Instance-Aware Test-Time Segmentation for Continual Domain Shifts
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.08569