1 paper
Jingyan Jiang, Yaru Sun, Xiao Chen +5
Test-time adaptation (TTA) can improve the recognition accuracy of vision-language models under distribution shift, but often degrades calibration, making predictive confidence unr…