ExDD: Explicit Dual Distribution Learning for Surface Defect Detection via Diffusion Synthesis

2025-07-21

AI Summary

Industrial anomaly detection faces two key challenges: severe scarcity of anomalous samples and the unrealistic single-class anomaly assumption—i.e., uniform anomaly distribution—which contradicts real-world industrial defect heterogeneity. To address these, we propose a dual-distribution modeling framework: (1) parallel memory banks explicitly encode normal and abnormal feature distributions; (2) a domain-text-guided latent diffusion model synthesizes high-fidelity, semantically grounded defect samples; and (3) a neighborhood-aware ratio scoring mechanism integrates multi-scale distance metrics. This work breaks from conventional single-class paradigms by enabling explicit, controllable modeling and synthesis of anomaly distributions for the first time. Evaluated on KSDD2, our method achieves 94.2% image-level AUROC (I-AUROC) and 97.7% pixel-level AUROC (P-AUROC), with peak performance attained using only 100 synthesized anomalies—significantly alleviating both data scarcity and distributional misspecification.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Computer Vision: Diffusion Models for Vision
• Machine Learning: Adversarial Learning & Robustness

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Industrial defect detection systems face critical limitations when confined to one-class anomaly detection paradigms, which assume uniform outlier distributions and struggle with data scarcity in realworld manufacturing environments. We present ExDD (Explicit Dual Distribution), a novel framework that transcends these limitations by explicitly modeling dual feature distributions. Our approach leverages parallel memory banks that capture the distinct statistical properties of both normality and anomalous patterns, addressing the fundamental flaw of uniform outlier assumptions. To overcome data scarcity, we employ latent diffusion models with domain-specific textual conditioning, generating in-distribution synthetic defects that preserve industrial context. Our neighborhood-aware ratio scoring mechanism elegantly fuses complementary distance metrics, amplifying signals in regions exhibiting both deviation from normality and similarity to known defect patterns. Experimental validation on KSDD2 demonstrates superior performance (94.2% I-AUROC, 97.7% P-AUROC), with optimal augmentation at 100 synthetic samples.

Problem

• Overcoming data scarcity in industrial defect detection
• Addressing flawed uniform outlier distribution assumptions
• Improving defect detection via synthetic data generation

Innovation

• Explicit dual distribution modeling for defects
• Diffusion synthesis with domain-specific conditioning
• Neighborhood-aware ratio scoring mechanism