Image Defect Detection Model Collection
List of Deep Learning models for defecting defect on an image dataset of metal assembly pieces
Object-detection models such as RetinaNet and CNN models finetuned via Transfer Learning to detect defects.
Fine-tuning pre-trained models via transfer learning is particularly effective when the target task has a limited amount of labeled data or when training from scratch is prohibitively expensive in terms of time or computational resources.
Defect detection, using an image dataset, is the process of analyzing images to identify and classify anomalies, imperfections, or irregularities within them, utilizing machine learning algorithms to automate the identification process and improve accuracy over time.
In this context several AI models have been implemented, from object-detection models such as RetinaNet to Convolutional Neural Networks pretrained and finetuned using open source datasets have been used. Those models are trained on different subsets of an image dataset provided by Bonfiglioli where metal pieces have been manually labelled.
Bonfiglioli is a company that provides full integrated solutions within a huge range of applications, from small gearmotors to big high-customized gearboxes, dedicated to all kind of customers and applications.
HLEG [102]
HLEG [102])