In the problem of out-of-distribution (OOD) detection, the usage of auxiliary
data as outlier data for fine-tuning has demonstrated encouraging performance.
However, previous methods have suffered from a trade-off between classification
accuracy (ACC) and OOD detection performance (AUR