Abstract:Existing anomaly detection paradigms overwhelmingly focus on training detection models using exclusively normal data or unlabeled data (mostly normal samples). One notorious issue with these approaches is that they are weak in discriminating anomalies from normal samples due to the lack of the knowledge about the anomalies. Here, we study the problem of few-shot anomaly detection, in which we aim at using a few labeled anomaly examples to train sample-efficient discriminative detection models. To address this problem, we introduce a novel weakly-supervised anomaly detection framework to train detection models without assuming the examples illustrating all possible classes of anomaly. Specifically, the proposed approach learns discriminative normality (regularity) by leveraging the labeled anomalies and a prior probability to enforce expressive representations of normality and unbounded deviated representations of abnormality. This is achieved by an end-to-end optimization of anomaly scores with a neural deviation learning, in which the anomaly scores of normal samples are imposed to approximate scalar scores drawn from the prior while that of anomaly examples is enforced to have statistically significant deviations from these sampled scores in the upper tail. Furthermore, our model is optimized to learn fine-grained normality and abnormality by top-K multiple-instance-learning-based feature subspace deviation learning, allowing more generalized representations. Comprehensive experiments on nine real-world image anomaly detection benchmarks show that our model is substantially more sample-efficient and robust, and performs significantly better than state-of-the-art competing methods in both closed-set and open-set settings. Our model can also offer explanation capability as a result of its prior-driven anomaly score learning. Code and datasets are available at: https://git.io/DevNet.

AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model

DualAnoDiff: Dual-Interrelated Diffusion Model for Few-Shot Anomaly Image Generation

Few-Shot Anomaly-Driven Generation for Anomaly Classification and Segmentation

A Diffusion-Based Framework for Multi-Class Anomaly Detection

AnomalySD: Few-Shot Multi-Class Anomaly Detection with Stable Diffusion Model

CAGEN: Controllable Anomaly Generator using Diffusion Model.

DiffusionAD: Norm-guided One-step Denoising Diffusion for Anomaly Detection

DiffusionAD: Denoising Diffusion for Anomaly Detection

Odd: One-Class Anomaly Detection Via the Diffusion Model

Unsupervised industrial anomaly detection with diffusion models

GLAD: Towards Better Reconstruction with Global and Local Adaptive Diffusion Models for Unsupervised Anomaly Detection

Unsupervised Surface Anomaly Detection with Diffusion Probabilistic Model

Image Anomaly Detection Based on Controllable Self-Augmentation

Self-supervised enhanced denoising diffusion for anomaly detection

Anomaly-GAN: A data augmentation method for train surface anomaly detection

A Novel Approach to Industrial Defect Generation through Blended Latent Diffusion Model with Online Adaptation

AnomalyXFusion: Multi-modal Anomaly Synthesis with Diffusion

Unsupervised Anomaly Detection via Masked Diffusion Posterior Sampling

AnomalyFactory: Regard Anomaly Generation as Unsupervised Anomaly Localization

Unseen Visual Anomaly Generation

Explainable Deep Few-shot Anomaly Detection with Deviation Networks