A Survey on Industrial Anomaly Synthesis
This survey reviews anomaly synthesis methods, introducing a new taxonomy covering about 40 techniques across four paradigms: Handcrafted, Distribution-hypothesis-based, Generative Model (GM)-based, and Vision Language Model (VLM)-based. It explores how these methods work to create realistic anomalies and points toward future research focusing on cross-modality synthesis and large VLM integration to improve anomaly generation quality.