Double Helix Diffusion for Cross-Domain Anomaly Image Generation

2025-09-16

AI Summary

In industrial visual anomaly detection, the scarcity of real anomalous samples and the structural inconsistency and feature entanglement inherent in existing synthetic methods pose critical bottlenecks. To address these challenges, we propose a dual-helix diffusion framework that jointly generates high-fidelity anomalous images and corresponding pixel-level masks. Our method employs domain-decoupled attention to disentangle normal and abnormal features, and introduces a semantic score map alignment module to ensure structural coherence and perceptual realism of generated anomalies. Furthermore, it supports fine-grained control via text and sketch guidance. Evaluated on benchmarks including MVTec-AD, our synthesized data achieves superior diversity and realism compared to state-of-the-art approaches. Downstream anomaly detection models trained on our synthetic data yield an average AUC improvement of 4.2%, demonstrating the framework’s effectiveness in alleviating annotation dependency and enhancing model generalization.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Web Mining and Content Analysis: Web data generation and simulation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Visual anomaly inspection is critical in manufacturing, yet hampered by the scarcity of real anomaly samples for training robust detectors. Synthetic data generation presents a viable strategy for data augmentation; however, current methods remain constrained by two principal limitations: 1) the generation of anomalies that are structurally inconsistent with the normal background, and 2) the presence of undesirable feature entanglement between synthesized images and their corresponding annotation masks, which undermines the perceptual realism of the output. This paper introduces Double Helix Diffusion (DH-Diff), a novel cross-domain generative framework designed to simultaneously synthesize high-fidelity anomaly images and their pixel-level annotation masks, explicitly addressing these challenges. DH-Diff employs a unique architecture inspired by a double helix, cycling through distinct modules for feature separation, connection, and merging. Specifically, a domain-decoupled attention mechanism mitigates feature entanglement by enhancing image and annotation features independently, and meanwhile a semantic score map alignment module ensures structural authenticity by coherently integrating anomaly foregrounds. DH-Diff offers flexible control via text prompts and optional graphical guidance. Extensive experiments demonstrate that DH-Diff significantly outperforms state-of-the-art methods in diversity and authenticity, leading to significant improvements in downstream anomaly detection performance.

Problem

• Generates synthetic anomaly images for training detectors
• Reduces feature entanglement between images and masks
• Ensures structural consistency of anomalies with background

Innovation

• Double Helix Diffusion cross-domain generative framework
• Domain-decoupled attention mechanism reduces feature entanglement
• Semantic score map alignment ensures structural authenticity