SOTAVerified

Out-of-Distribution Detection Without Class Labels

2021-12-14Unverified0· sign in to hype

Niv Cohen, Ron Abutbul, Yedid Hoshen

Unverified — Be the first to reproduce this paper.

Reproduce

Abstract

Out-of-distribution detection seeks to identify novelties, samples that deviate from the norm. The task has been found to be quite challenging, particularly in the case where the normal data distribution consists of multiple semantic classes (e.g., multiple object categories). To overcome this challenge, current approaches require manual labeling of the normal images provided during training. In this work, we tackle multi-class novelty detection without class labels. Our simple but effective solution consists of two stages: we first discover "pseudo-class" labels using unsupervised clustering. Then using these pseudo-class labels, we are able to use standard supervised out-of-distribution detection methods. We verify the performance of our method by a favorable comparison to the state-of-the-art, and provide extensive analysis and ablations.

Tasks

Benchmark Results

DatasetModelMetricClaimedVerifiedStatus
Anomaly Detection on Anomaly Detection on Unlabeled ImageNet-30 vs Flowers-102PsudoLabels CLIP ViTROC-AUC98.3Unverified
Anomaly Detection on Unlabeled CIFAR-10 vs LSUN (Fix)PsudoLabels ViTROC-AUC99.1Unverified
Anomaly Detection on Unlabeled CIFAR-10 vs LSUN (Fix)PsudoLabels ResNet-152ROC-AUC95.7Unverified
Anomaly Detection on Unlabeled CIFAR-10 vs LSUN (Fix)PsudoLabels ResNet-18ROC-AUC94.3Unverified
Anomaly Detection on Unlabeled CIFAR-10 vs LSUN (Fix)SCAN FeaturesROC-AUC90.2Unverified
Anomaly Detection on Unlabeled ImageNet-30 vs CUB-200PsudoLabels CLIP ViTROC-AUC99.4Unverified
Unlabeled CIFAR-10 vs CIFAR-100PsudoLabels ResNet-152AUROC93.3Unverified
Unlabeled CIFAR-10 vs CIFAR-100PsudoLabels ResNet-18AUROC90.8Unverified
Unlabeled CIFAR-10 vs CIFAR-100SCAN FeaturesAUROC90.2Unverified
Unlabeled CIFAR-10 vs CIFAR-100PsudoLabels ViTAUROC96.7Unverified

Reproductions