Continual Test-Time Domain Adaptation

By Qin Wang, Olga Fink, Luc Van Gool, Dengxin Dai. 2022

754 citations recorded by the source corpus.

Test-time domain adaptation aims to adapt a source pre-trained model to a target domain without using any source data. Existing works mainly consider the case where the target domain is static. However, real-world machine perception systems are running in non-stationary and continually changing environments where the target domain distribution can change over time. Existing methods, which are mostly based on self-training and entropy regularization, can suffer from these non-stationary environments. Due to the distribution shift over time in the target domain, pseudo-labels become unreliable. The noisy pseudo-labels can further lead to error accumulation and catastrophic forgetting. To tackle these issues, we propose a continual test-time adaptation approach (CoTTA) which comprises two parts. Firstly, we propose to reduce the error accumulation by using weight-averaged and augmentation-averaged predictions which are often more accurate. On the other hand, to avoid catastrophic forgetting, we propose to stochastically restore a small part of the neurons to the source pre-trained weights during each iteration to help preserve source knowledge in the longterm. The proposed method enables the longterm adaptation for all parameters in the network. CoTTA is easy to implement and can be readily incorporated in off-the-shelf pre-trained models. We demonstrate the effectiveness of our approach on four classification tasks and a segmentation task for continual test-time adaptation, on which we outperform existing methods. Our code is available at https://gin.ee/cotta.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Towards Label-Efficient Incremental Learning: A Survey (2023, background): “Continual Test-Time Adaptation [ Wang et al. , 2022 ] proposes to adapt a pre-trained deep classiﬁer at inference time.”
Recent Advances of Continual Learning in Computer Vision: An Overview (2021, neutral): “Specifically, FGN designs a new graph topology through which the continual growth of a graph can be regarded as the continual increase of training samples, and typical convolutional-neural-network-based continual learning techniques can thus be used for graph learning.”
NOTE: Robust Continual Test-time Adaptation Against Temporal Correlation (2022, methodology): “They are temporally correlated. existing TTA algorithms simply use an incoming batch of test samples for adaptation [29, 33, 41, 44], the model might be biased towards these imbalanced samples under the temporally correlated test streams.”
ViDA: Homeostatic Visual Domain Adapter for Continual Test Time Adaptation (2023, methodology): “To tackle this problem, Continual Test-Time Adaptation (CTTA) has been proposed (Wang et al., 2022), which addresses a sequence of different distribution shifts over time rather than a single shift as in TTA.”
HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (2022, neutral): “Therefore, several methods aim to reduce the annotation burden such as weakly-supervised learning [14, 61, 93], semi-supervised learning [17, 27, 35, 53], and unsupervised domain adaption (UDA) [26, 28, 59, 67, 92].”
Deep Class-Incremental Learning: A Survey (2023, background): “Besides, test-time adaptation [219] enables the model to handle these drifts on-the-fly.”
Efficient Test-Time Model Adaptation without Forgetting (2022, background): “Recently, several attempts (Sun et al., 2020; Wang et al., 2021; Liu et al., 2021; Zhang et al., 2021b;c; Wang et al., 2022) have been proposed to handle the distribution shifts by online adapting a model at test time (called test-time adaptation ).”
Towards Stable Test-Time Adaptation in Dynamic Wild World (2023, background): “…et al., 2021; Lim et al., 2023; Zhao et al., 2023), unsupervised entropy minimization (Wang et al., 2021; Niu et al., 2022a; Zhang et al., 2022), prediction consistency maximization (Zhang et al., 2022; Wang et al., 2022; Chen et al., 2022a), top-k classification boosting (Niu et al., 2022b), etc.”
DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation (2021, methodology): “Several methods also combine adversarial and self-training [37,40,74], train with auxiliary tasks [32, 73, 75], or perform test-time UDA [76].”