Exposing Outlier Exposure: What Can Be Learned From Few, One, and Zero Outlier Images

Open paper
Benchmark Model Rank Results
anomaly-detection-on-leave-one-class-out-cifar-10BCE-CLIP#1AUROC: 98.4
anomaly-detection-on-leave-one-class-out-cifar-10CLIP (zero shot)#2AUROC: 92.2
anomaly-detection-on-leave-one-class-out-cifar-10Binary Cross Entropy (OE)#3AUROC: 86.6
anomaly-detection-on-leave-one-class-out-cifar-10HSC#4AUROC: 84.8
anomaly-detection-on-leave-one-class-out-cifar-10DSAD#5AUROC: 84.2
anomaly-detection-on-leave-one-class-out-cifar-10DSVDD#6AUROC: 52.2
anomaly-detection-on-leave-one-class-out-imagenet-30BCE-CLIP (OE)#1AUROC: 99.3
anomaly-detection-on-leave-one-class-out-imagenet-30CLIP (zero shot)#2AUROC: 97.8
anomaly-detection-on-leave-one-class-out-imagenet-30DSAD#3AUROC: 88.8
anomaly-detection-on-leave-one-class-out-imagenet-30HSC (OE)#4AUROC: 88.3
anomaly-detection-on-leave-one-class-out-imagenet-30Binary Cross Entropy (OE)#5AUROC: 88.2
anomaly-detection-on-leave-one-class-out-imagenet-30DSVDD#6AUROC: 49.7
anomaly-detection-on-one-class-cifar-10CLIP (OE)#1AUROC: 99.6
anomaly-detection-on-one-class-cifar-10CLIP (zero shot)#6AUROC: 98.5
anomaly-detection-on-one-class-imagenet-30BCE-Clip (OE)#1AUROC: 99.90
anomaly-detection-on-one-class-imagenet-30CLIP (Zero Shot)#2AUROC: 99.88
anomaly-detection-on-one-class-imagenet-30Binary Cross Entropy (OE)#3AUROC: 97.7

Rank counts only results with a code link.