TeSLA: Test-Time Self-Learning With Automatic Adversarial Augmentation
This addresses the need for robust and efficient adaptation of models during deployment, particularly in dynamic or domain-shifted environments like medical imaging, though it appears incremental by building on existing self-learning and adversarial augmentation concepts.
The paper tackled the problem of test-time adaptation for pre-trained models on unlabeled streaming data by proposing TeSLA, a method that uses a novel loss function and automatic adversarial augmentation, achieving state-of-the-art results in classification and segmentation across various domain shifts, including challenging medical image shifts.
Most recent test-time adaptation methods focus on only classification tasks, use specialized network architectures, destroy model calibration or rely on lightweight information from the source domain. To tackle these issues, this paper proposes a novel Test-time Self-Learning method with automatic Adversarial augmentation dubbed TeSLA for adapting a pre-trained source model to the unlabeled streaming test data. In contrast to conventional self-learning methods based on cross-entropy, we introduce a new test-time loss function through an implicitly tight connection with the mutual information and online knowledge distillation. Furthermore, we propose a learnable efficient adversarial augmentation module that further enhances online knowledge distillation by simulating high entropy augmented images. Our method achieves state-of-the-art classification and segmentation results on several benchmarks and types of domain shifts, particularly on challenging measurement shifts of medical images. TeSLA also benefits from several desirable properties compared to competing methods in terms of calibration, uncertainty metrics, insensitivity to model architectures, and source training strategies, all supported by extensive ablations. Our code and models are available on GitHub.