DPCore: Dynamic Prompt Coreset for Continual Test-Time Adaptation

Continual Test-Time Adaptation (CTTA) faces challenges when real-world domains are dynamic-recurring with varying frequencies and durations-unlike the structured changes many methods assume. Existing approaches then struggle with convergence issues from brief domain exposures, catastrophic forgetting, and knowledge misapplication in these dynamic conditions. We propose DPCore, a robust and computationally efficient method designed for such dynamic patterns. DPCore integrates three key components: Visual Prompt Adaptation for efficient domain alignment, a Prompt Coreset for knowledge preservation, and a Dynamic Update mechanism that intelligently manages prompts based on domain similarity. Extensive experiments on four benchmarks show DPCore achieves state-of-theart performance in both structured and dynamic settings, significantly reducing trainable parameters by 99% and computation time by 64% compared to previous approaches.