Abstract:In industrial scenarios, it is crucial not only to identify anomalous items but also to classify the type of anomaly. However, research on anomaly multi-classification remains largely unexplored. This paper proposes a novel and valuable research task called anomaly multi-classification. Given the challenges in applying few-shot learning to this task, due to limited training data and unique characteristics of anomaly images, we introduce a baseline model that combines RelationNet and PatchCore. We propose a data generation method that creates pseudo classes and a corresponding proxy task, aiming to bridge the gap in transferring few-shot learning to industrial scenarios. Furthermore, we utilize contrastive learning to improve the vanilla baseline, achieving much better performance than directly fine-tune a ResNet. Experiments conducted on MvTec AD and MvTec3D AD demonstrate that our approach shows superior performance in this novel task.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is not only to identify abnormal items in industrial scenarios, but also to classify the types of abnormalities. However, there are relatively few studies on multi - class anomaly classification. The paper proposes a new and valuable research task - multi - class anomaly classification, and in view of the challenges faced by applying few - shot learning in this task, such as limited training data and the unique characteristics of abnormal images, the authors introduce a baseline model that combines RelationNet and PatchCore. In addition, they propose a data generation method for creating pseudo - classes and corresponding proxy tasks, aiming to bridge the gap in transferring few - shot learning to industrial scenarios. By using contrastive learning to improve the base model, the experiments of the authors' method on the MvTec AD and MvTec3D AD datasets show superior performance in this novel task. Specifically, the main contributions of the paper include: 1. **Establishing a baseline model**: A base model is established for the newly proposed task - few - shot multi - class anomaly classification based on the RelationNet network. 2. **Data generation method**: A data generation method is proposed, and these data are used to pre - train the network to complete proxy tasks, verifying the effectiveness of data generation and proxy tasks in transferring few - shot learning to anomaly classification tasks, thus solving the problem of lack of training sets. 3. **Residual representation**: By using the residual features extracted by PatchCore instead of the original image features, the classification performance is improved, making the base model more suitable for industrial anomaly classification tasks. 4. **Improvement of contrastive learning**: Inspired by contrastive learning, the base model is modified to further improve the classification performance. Through these methods, the authors effectively solve the challenges of multi - class anomaly classification in industrial scenarios, especially in the case of a limited number of samples.

Anomaly Multi-classification in Industrial Scenarios: Transferring Few-shot Learning to a New Task

Learning Discrimination from Contaminated Data: Multi-Instance Learning for Unsupervised Anomaly Detection

AnomalyNCD: Towards Novel Anomaly Class Discovery in Industrial Scenarios

AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model

Context-aware Feature Reconstruction for Class-Incremental Anomaly Detection and Localization

A Diffusion-Based Framework for Multi-Class Anomaly Detection

Focus Your Distribution: Coarse-to-Fine Non-Contrastive Learning for Anomaly Detection and Localization

A Two-Stage Anomaly Detection Framework: Towards Low Omission Rate in Industrial Vision Applications.

Prioritized Local Matching Network for Cross-Category Few-Shot Anomaly Detection

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

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

DMMGNet: A Discrimination Mapping and Memory Bank Mean Guidance-based Network for High-performance Few-shot Industrial Anomaly Detection

Anomaly-prior Guided Inpainting for Industrial Visual Anomaly Detection

CLIP-FSAC++: Few-Shot Anomaly Classification with Anomaly Descriptor Based on CLIP

TF2: Few-shot Text-Free Training-Free Defect Image Generation for Industrial Anomaly Inspection

Towards Zero-shot 3D Anomaly Localization

AnomalyDINO: Boosting Patch-based Few-shot Anomaly Detection with DINOv2

Multimodal Industrial Anomaly Detection via Hybrid Fusion

Mixed Noise-Guided Mutual Constraint Framework for Unsupervised Anomaly Detection in Smart Industries

PTMNet: Pixel-Text Matching Network for Zero-Shot Anomaly Detection

Multi-Scale Memory Comparison for Zero-/Few-Shot Anomaly Detection