STAR: Test-Time Adaptation Can Enhance Universal Prompt Learning for Vision-Language Models

Benchmark Model Rank Results
out-of-distribution-detection-on-imagenet-1k-vs-inaturalist-ood-mos-mcm-full-50kSTAR L (16-shot)–AUROC: 99.14FPR95: 3.85
out-of-distribution-detection-on-imagenet-1k-vs-sun-ood-mos-mcm-full-50kSTAR L (16-shot)–AUROC: 97.23FPR95: 13.94
out-of-distribution-detection-on-imagenet-1k-vs-textures-ood-mos-protocolSTAR L (16-shot)–AUROC: 89.98FPR95: 41.54