Jihoon Tack, Sangwoo Mo, Jongheon Jeong, Jinwoo Shin
Novelty detection, i.e., identifying whether a given sample is drawn from outside the training distribution, is essential for reliable machine learning. To this end, there have been many attempts at learning a representation well-suited for novelty detection and designing a score based on such representation. In this paper, we propose a simple, yet effective method named contrasting shifted instances (CSI), inspired by the recent success on contrastive learning of visual representations. Specifically, in addition to contrasting a given sample with other instances as in conventional contrastive learning methods, our training scheme contrasts the sample with distributionally-shifted augmentations of itself. Based on this, we propose a new detection score that is specific to the proposed training scheme. Our experiments demonstrate the superiority of our method under various novelty detection scenarios, including unlabeled one-class, unlabeled multi-class and labeled multi-class settings, with various image benchmark datasets. Code and pre-trained models are available at https://github.com/alinlab/CSI.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Anomaly Detection | One-class ImageNet-30 | AUROC | 91.6 | CSI |
| Anomaly Detection | Anomaly Detection on Anomaly Detection on Unlabeled ImageNet-30 vs Flowers-102 | ROC-AUC | 94.7 | CSI |
| Anomaly Detection | Anomaly Detection on Unlabeled CIFAR-10 vs LSUN (Fix) | ROC-AUC | 90.3 | CSI |
| Anomaly Detection | Unlabeled CIFAR-10 vs CIFAR-100 | AUROC | 89.3 | CSI |
| Anomaly Detection | One-class CIFAR-100 | AUROC | 89.6 | CSI |
| Anomaly Detection | Anomaly Detection on Unlabeled ImageNet-30 vs CUB-200 | ROC-AUC | 71.5 | CSI |
| Anomaly Detection | One-class CIFAR-10 | AUROC | 94.3 | CSI |