Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations

Benchmark Model Rank Results
image-classification-on-imagenetCMA(ViT-B/16)#501Top 1 Accuracy: 82.64%
out-of-distribution-detection-on-imagenet-1k-10CMA(ViT-B/16, NegLabel)#14AUROC: 91.64FPR95: 33.58
out-of-distribution-detection-on-imagenet-1k-3CMA(ViT-B/16, NegLabel)#2AUROC: 99.62FPR95: 1.65
out-of-distribution-detection-on-imagenet-1k-8CMA(ViT-B/16, NegLabel)#1FPR95: 16.84AUROC: 96.36
out-of-distribution-detection-on-imagenet-1k-9CMA(ViT-B/16, NegLabel)#1FPR95: 27.65AUROC: 93.11