ProGAN generates realistic images to improve metal defect detection
ProGAN-Based Synthetic Image Generation for Industrial Visual Inspection of Metallic Parts
Computer Vision and Pattern Recognition
Summary
Detecting defects in metal parts helps make repairs and recycling work better, but computers need lots of images to learn what defects look like. The authors show how a special AI called ProGAN can create realistic fake images of metal parts to help train these computers. They tested their method on two types of metal images and found it made adding new images stable and realistic. This means better training for defect detection without needing to collect many rare damaged samples.
What this means in practice
- •For industrial inspection teams: Enhance training datasets with synthetic images to improve defect detection models for metallic parts in manufacturing.
- •For recycling facility operators: Use AI-trained models supported by synthetic image data to better identify metal defects for sorting and recycling processes.
Authors
Hossein Omid Beiki, Sepehr Asgari, Oliver Petrovic, Christian Brecher
Abstract
Automated visual inspection is essential for efficient repair and recycling of metallic components. AI-based defect detection is a strong fit. However, it is often constrained by scarce, low-diversity training data. To address this gap, we study Generative AI-driven synthetic image generation using Progressive Growing GANs (ProGAN). We implement a ProGAN pipeline with Wasserstein loss and gradient penalty and evaluate key hyperparameters on two public datasets, MVTec AD (metal nut) and MPDD (tube). As a result, training remained stable and the model yielded realistic images. These images can augment training data, improve robustness, and reduce the cost of collecting rare defects.