Abstract: The robustness of deep learning model can be compromised by out-of-distribution (OOD) testing data. Test-time adaptation (TTA) emerges as an efficient method to mitigate the distribution gap ...