Test-time Adaptation against Multi-modal Reliability Bias

Background: Distribution Shifts Out of distribution (OOD): the pre-trained models always encounter performance degeneration when the distribution shifts between training and test data emerge in the scenarios of • Changing weather, e.g., fog, snow • Degenerated sensors, e.g. defocus, gaussian noise Results under Mixed Severity Levels Contact Information Feel free to drop me an email.