Anomaly Detection via Reverse Distillation from One-Class Embedding

CVPR 2022  ·  Hanqiu Deng, Xingyu Li ·

Knowledge distillation (KD) achieves promising results on the challenging problem of unsupervised anomaly detection (AD).The representation discrepancy of anomalies in the teacher-student (T-S) model provides essential evidence for AD. However, using similar or identical architectures to build the teacher and student models in previous studies hinders the diversity of anomalous representations. To tackle this problem, we propose a novel T-S model consisting of a teacher encoder and a student decoder and introduce a simple yet effective "reverse distillation" paradigm accordingly. Instead of receiving raw images directly, the student network takes teacher model's one-class embedding as input and targets to restore the teacher's multiscale representations. Inherently, knowledge distillation in this study starts from abstract, high-level presentations to low-level features. In addition, we introduce a trainable one-class bottleneck embedding (OCBE) module in our T-S model. The obtained compact embedding effectively preserves essential information on normal patterns, but abandons anomaly perturbations. Extensive experimentation on AD and one-class novelty detection benchmarks shows that our method surpasses SOTA performance, demonstrating our proposed approach's effectiveness and generalizability.

PDF Abstract CVPR 2022 PDF CVPR 2022 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Anomaly Detection AeBAD-S ReverseDistillation Segmentation AUPRO 85.6 # 3
Detection AUROC 81.0 # 2
Anomaly Detection AeBAD-V ReverseDistillation Detection AUROC 71.0 # 2
Anomaly Detection Fashion-MNIST Reverse Distillation ROC AUC 95.0 # 3
Anomaly Detection MVTec AD Reverse Distillation Detection AUROC 98.5 # 37
Segmentation AUROC 97.8 # 42
Segmentation AUPRO 93.9 # 22
Anomaly Detection MVTec LOCO AD RD4AD Avg. Detection AUROC 78.7 # 25
Detection AUROC (only logical) 69.4 # 29
Detection AUROC (only structural) 88.0 # 13
Segmentation AU-sPRO (until FPR 5%) 63.7 # 9
Anomaly Detection One-class CIFAR-10 Reverse Distillation AUROC 86.5 # 22
Anomaly Detection VisA Reverse Distillation Segmentation AUPRO (until 30% FPR) 70.9 # 16

Methods