Anomaly Detection with Synthetic Images
Research Paper#Anomaly Detection, Synthetic Data, Image Generation🔬 Research|Analyzed: Jan 3, 2026 19:05•
Published: Dec 29, 2025 06:06
•1 min read
•ArXivAnalysis
This paper addresses the challenge of anomaly detection in industrial manufacturing, where real defect images are scarce. It proposes a novel framework to generate high-quality synthetic defect images by combining a text-guided image-to-image translation model and an image retrieval model. The two-stage training strategy further enhances performance by leveraging both rule-based and generative model-based synthesis. This approach offers a cost-effective solution to improve anomaly detection accuracy.
Key Takeaways
- •Addresses the scarcity of real defect images in industrial anomaly detection.
- •Proposes a framework using text-guided image-to-image translation and image retrieval for synthetic defect image generation.
- •Employs a two-stage training strategy to leverage both rule-based and generative synthesis.
- •Demonstrates effectiveness on the MVTec AD dataset.
Reference / Citation
View Original"The paper introduces a novel framework that leverages a pre-trained text-guided image-to-image translation model and image retrieval model to efficiently generate synthetic defect images."