Distill Once, Adapt Life-Long: Exploring Dataset Distillation for Continual Test-Time Adaptation
For practitioners deploying models under evolving target domains with privacy constraints, DO-ALL offers a plug-and-play solution to mitigate catastrophic forgetting and compounding errors in source-free CTTA.
DO-ALL uses dataset distillation to generate synthetic anchors from the source domain, enabling stable continual test-time adaptation without retaining raw source data, and consistently improves long-term robustness across CIFAR100-C, ImageNet-C, and CCC benchmarks.
Continual Test-Time Adaptation (CTTA) aims to maintain model performance under evolving target domains by adapting online without labeled data. However, practical deployments often cannot retain the source dataset due to privacy or licensing constraints, and purely source-free CTTA methods tend to become unstable under long-term distribution shift, suffering from compounding self-training errors and catastrophic forgetting. We introduce DO-ALL (Distill Once, Adapt Life-Long), a plug-and-play framework that revisits source information in a compact and privacy-conscious form via Dataset Distillation (DD). Before deployment, DO-ALL performs DD to produce a small set of synthetic distilled anchors that summarize the source distribution. During adaptation, each target sample is matched with its most semantically aligned anchor, which provides a stable reference for various CTTA via source replay, representation alignment, and manifold-smoothing regularization. DO-ALL can be seamlessly integrated into existing CTTA algorithms, consistently improving long-term robustness across CIFAR100-C, ImageNet-C, and the CCC benchmark. This demonstrates the potential of leveraging DD to enable stable and continuous adaptation without retaining raw source data. The code is available at https://github.com/blue-531/DOALL.