Learning to Adapt to Online Streams with Distribution Shifts
This work addresses a practical challenge in test-time adaptation for real-world applications like video segmentation, though it is incremental as it builds on existing TTA techniques.
The authors tackled the problem of adapting models to online data streams with distribution shifts by proposing a meta-learning approach that enables continual adaptation without batch size restrictions, achieving consistent improvements over state-of-the-art methods in benchmarking datasets and superior performance in video segmentation.
Test-time adaptation (TTA) is a technique used to reduce distribution gaps between the training and testing sets by leveraging unlabeled test data during inference. In this work, we expand TTA to a more practical scenario, where the test data comes in the form of online streams that experience distribution shifts over time. Existing approaches face two challenges: reliance on a large test data batch from the same domain and the absence of explicitly modeling the continual distribution evolution process. To address both challenges, we propose a meta-learning approach that teaches the network to adapt to distribution-shifting online streams during meta-training. As a result, the trained model can perform continual adaptation to distribution shifts in testing, regardless of the batch size restriction, as it has learned during training. We conducted extensive experiments on benchmarking datasets for TTA, incorporating a broad range of online distribution-shifting settings. Our results showed consistent improvements over state-of-the-art methods, indicating the effectiveness of our approach. In addition, we achieved superior performance in the video segmentation task, highlighting the potential of our method for real-world applications.