CD-Buffer: Complementary Dual-Buffer Framework for Test-Time Adaptation in Adverse Weather Object Detection

Abstract

Test-Time Adaptation (TTA) enables real-time adaptation to domain shifts without off-line retraining. Recent TTA methods have predominantly explored additive approaches that introduce lightweight modules for feature refinement. Recently, a subtractive approach that removes domain-sensitive channels has emerged as an alternative direction. We observe that these paradigms exhibit complementary effectiveness patterns: subtractive methods excel under severe shifts by removing corrupted features, while additive methods are effective under moderate shifts requiring refinement. However, each paradigm operates effectively only within limited shift severity ranges, failing to generalize across diverse corruption levels. This leads to the following question: can we adaptively balance both strategies based on measured feature-level domain shift? We propose CD-Buffer, a novel complementary dual-buffer framework where subtractive and additive mechanisms operate in opposite yet coordinated directions driven by a unified discrepancy metric. Our key innovation lies in the discrepancy-driven coupling: Our framework couples removal and refinement through a unified discrepancy metric, automatically balancing both strategies based on feature-level shift severity. This establishes automatic channel-wise balancing that adapts differentiated treatment to heterogeneous shift magnitudes without manual tuning. Extensive experiments on KITTI, Cityscapes, and ACDC datasets demonstrate state-of-the-art performance, consistently achieving superior results across diverse weather conditions and severity levels.

Figures (8)

• Figure 1: Adaptation Performance Across Fog Severities. Additive method kim2025buffer excels under moderate shifts but struggle under severe corruption, while subtractive method wang2025pruning shows the opposite pattern. Our CD-Buffer achieves consistent performance across all severities by adaptively balancing both strategies.
• Figure 2: Continual test-time adaptation (TTA) performance on KITTI Fog across different severities. The plots show the evolution of mAP@50 over adaptation steps for (a) 50m, (b) 75m, and (c) 150m fog distances. Our method (blue) achieves the fastest and most stable improvement, maintaining superior accuracy throughout adaptation.
• Figure 3: Effect of key hyperparameters on TTA performance:(a) regularization weight $\lambda_{\mathrm{reg}}$ and (b) pruning ratio $\rho_{\text{target}}$.
• Figure 4: Visualization of continual TTA process on KITTI fog with cyclic severity transitions (50m $\rightarrow$ 75m $\rightarrow$ 150m, repeated 10 times). Our method maintains consistently superior performance across all cycles, demonstrating both rapid adaptation within each severity level and robustness to various severity.
• Figure 5: Discrepancy across fog severities and backbone stages.