CORE: Contrast-and-Reflect for Self-Supervised Prompt Optimization in Anomaly Synthesis

Key points

• Industrial anomaly detection (AD) is bottlenecked by the scarcity of anomalous samples, making high-fidelity anomaly synthesis crucial for improving downstream performance.1
• However, existing methods either rely on scarce real anomaly references or auxiliary data, while lacking complex reasoning and iterative optimization capabilities for generating realistic defects in open-set scenarios.1
• CORE first introduces an Anomaly Context Grounding module to transform sparse user context into defect-aware generation context, which is subsequently instantiated as structured anomaly prompts that disentangle normal-region preservation from anomaly-specific variation, enabling controllable and faithful defect generation.1
• Extensive experiments on MVTec-AD and VisA demonstrate that CORE synthesizes realistic and diverse anomalies, yielding superior improvements in downstream AD performance.1

Sources (1)

• [1]CORE: Contrast-and-Reflect for Self-Supervised Prompt Optimization in Anomaly Synthesis
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 05:33 AM

  To address these challenges, we propose CORE, a cost-efficient and training-free prompt optimization framework based on Large Multimodal Models (LMMs), which unifies zero-shot and one-shot anomaly synthesis.

  Industrial anomaly detection (AD) is bottlenecked by the scarcity of anomalous samples, making high-fidelity anomaly synthesis crucial for improving downstream performance.

Extractive summary: sentences quoted from the sources.