<ul><li>This paper introduces a self-bootstrapping scheme for versatile test-time adaptation (TTA) objective in various tasks including image classification, regression, object-level predictions, and pixel-level predictions.</li><li>The scheme optimizes prediction consistency between the test image and its deteriorated view.</li><li>The paper addresses challenges related to preserving geometric information and providing sufficient learning signals for TTA.</li><li>Experiments demonstrate superior results of the proposed method across different tasks.</li></ul>

Self-Bootstrapping for Versatile Test-Time Adaptation

Discover more