Abstract:Since autonomous driving systems usually face dynamic and ever-changing environments, continual test-time adaptation (CTTA) has been proposed as a strategy for transferring deployed models to continually changing target domains. However, the pursuit of long-term adaptation often introduces catastrophic forgetting and error accumulation problems, which impede the practical implementation of CTTA in the real world. Recently, existing CTTA methods mainly focus on utilizing a majority of parameters to fit target domain knowledge through self-training. Unfortunately, these approaches often amplify the challenge of error accumulation due to noisy pseudo-labels, and pose practical limitations stemming from the heavy computational costs associated with entire model updates. In this paper, we propose a distribution-aware tuning (DAT) method to make the semantic segmentation CTTA efficient and practical in real-world applications. DAT adaptively selects and updates two small groups of trainable parameters based on data distribution during the continual adaptation process, including domain-specific parameters (DSP) and task-relevant parameters (TRP). Specifically, DSP exhibits sensitivity to outputs with substantial distribution shifts, effectively mitigating the problem of error accumulation. In contrast, TRP are allocated to positions that are responsive to outputs with minor distribution shifts, which are fine-tuned to avoid the catastrophic forgetting problem. In addition, since CTTA is a temporal task, we introduce the Parameter Accumulation Update (PAU) strategy to collect the updated DSP and TRP in target domain sequences. We conduct extensive experiments on two widely-used semantic segmentation CTTA benchmarks, achieving promising performance compared to previous state-of-the-art methods.

Improving cross-domain generalizability of medical image segmentation using uncertainty and shape-aware continual test-time domain adaptation

Multi Task Consistency Guided Source-Free Test-Time Domain Adaptation Medical Image Segmentation

Multi-Modal Continual Test-Time Adaptation for 3D Semantic Segmentation

A Versatile Framework for Continual Test-Time Domain Adaptation: Balancing Discriminability and Generalizability

A Prototypical Knowledge Oriented Adaptation Framework for Semantic Segmentation

SATTA: Semantic-Aware Test-Time Adaptation for Cross-Domain Medical Image Segmentation

Test-Time Adaptation with Shape Moments for Image Segmentation

Bridge Segmentation Performance Gap Via Evolving Shape Prior.

DG-TTA: Out-of-domain medical image segmentation through Domain Generalization and Test-Time Adaptation

Continuous Adaptation for Interactive Segmentation Using Teacher-Student Architecture

Uncertainty-Aware Consistency Regularization for Cross-Domain Semantic Segmentation

Distribution-Aware Continual Test-Time Adaptation for Semantic Segmentation

Fully Test-Time Adaptation for Image Segmentation

CAT-DG: A Cross-Attention-Based Domain Generalization Model for Medical Image Segmentation

CyCMIS: Cycle-consistent Cross-domain Medical Image Segmentation Via Diverse Image Augmentation

Edge-Guided Output Adaptor: Highly Efficient Adaptation Module for Cross-Vendor Medical Image Segmentation.

Learning Instance-Specific Adaptation for Cross-Domain Segmentation

Test-time adaptation for geospatial point cloud semantic segmentation with distinct domain shifts

Towards collaborative appearance and semantic adaptation for medical image segmentation

Unsupervised Domain Adaptation for Medical Image Segmentation by Selective Entropy Constraints and Adaptive Semantic Alignment

Tri-Directional Tasks Complementary Learning for Unsupervised Domain Adaptation of Cross-modality Medical Image Semantic Segmentation.