Large-Scale Universal Defect Generation: Foundation Models and Datasets

TL;DR

This paper introduces UDG, a large-scale defect dataset, and UniDG, a universal foundation model for defect generation that improves diversity, realism, and generalization without per-category fine-tuning.

Contribution

The work presents a novel large-scale defect dataset and a versatile foundation model capable of reference-based and instruction-based defect editing across diverse categories.

Findings

UniDG outperforms prior methods in synthesis quality.

Extensive experiments show improved anomaly detection and localization.

The dataset enables better generalization across defect categories.

Abstract

Existing defect/anomaly generation methods often rely on few-shot learning, which overfits to specific defect categories due to the lack of large-scale paired defect editing data. This issue is aggravated by substantial variations in defect scale and morphology, resulting in limited generalization, degraded realism, and category consistency. We address these challenges by introducing UDG, a large-scale dataset of 300K normal-abnormal-mask-caption quadruplets spanning diverse domains, and by presenting UniDG, a universal defect generation foundation model that supports both reference-based defect generation and text instruction-based defect editing without per-category fine-tuning. UniDG performs Defect-Context Editing via adaptive defect cropping and structured diptych input format, and fuses reference and target conditions through MM-DiT multimodal attention. A two-stage training strategy, Diversity-SFT followed by Consistency-RFT, further improves diversity while enhancing realism and reference consistency. Extensive experiments on MVTec-AD and VisA show that UniDG outperforms prior few-shot anomaly generation and image insertion/editing baselines in synthesis quality and downstream single- and multi-class anomaly detection/localization. Code will be available at https://github.com/RetoFan233/UniDG.