A Survey on Industrial Anomalies Synthesis

TL;DR

Industrial anomaly detection in manufacturing suffers from scarce defective samples and costly annotation. This survey provides a unified IAS taxonomy across four paradigms—hand-crafted, distribution-hypothesis-based, GM-based, and VLM-based synthesis—and analyzes cross-modality strategies and large-scale VLM integration. It covers roughly 40 representative methods, contrasts strengths and limitations, and offers a practical roadmap to leverage multimodal cues to improve downstream detection. The work aims to guide researchers and practitioners in selecting synthesis strategies aligned with real-world data constraints and industrial needs.

Abstract

This paper comprehensively reviews anomaly synthesis methodologies. Existing surveys focus on limited techniques, missing an overall field view and understanding method interconnections. In contrast, our study offers a unified review, covering about 40 representative methods across Hand-crafted, Distribution-hypothesis-based, Generative models (GM)-based, and Vision-language models (VLM)-based synthesis. We introduce the first industrial anomaly synthesis (IAS) taxonomy. Prior works lack formal classification or use simplistic taxonomies, hampering structured comparisons and trend identification. Our taxonomy provides a fine-grained framework reflecting methodological progress and practical implications, grounding future research. Furthermore, we explore cross-modality synthesis and large-scale VLM. Previous surveys overlooked multimodal data and VLM in anomaly synthesis, limiting insights into their advantages. Our survey analyzes their integration, benefits, challenges, and prospects, offering a roadmap to boost IAS with multimodal learning. More resources are available at https://github.com/M-3LAB/awesome-anomaly-synthesis.

Figures (2)

• Figure 2: A Taxonomy of Image Anomalies Synthesis(IAS)
• Figure 3: Illustration of different IAS methods. It categorizes IAS methods into four main approaches: (a) Hand-crafted synthesis, which relies on pre-defined textures such as external-dependent synthesis, self-contained synthesis, and inpainting-based synthesis. (b) Distribution hypothesis-based synthesis, which synthesizes anomalies by perturbing learned feature distributions of normal data through prior-dependent or data-driven synthesis. (c) Generative models (GM)-based synthesis, which employs generative models to perform full-image generation, full-image translation, or local anomaly synthesis. (d) Vision-language models (VLM)-based synthesis, which integrates textual features into a single-stage or multi-stage synthesis process.