Neurocle has launched version 4.0 of Neuro-T, a deep learning trainer that creates AI models for visual inspection, and Neuro-R, a runtime API that applies the inspection model created by Neuro-T to the manufacturing production line in real time. The new versions of the software address the problems of insufficient images of defects for training neural networks and finite resources for labeling. It also includes several other new features.
Training deep learning models requires images of both normal and defective products, but it can be difficult to collect enough images of product defects. Neuro-T now includes an AI model for generating virtual defects, or GAN, which allows users to generate synthetic data for product defects.
Version 4.0 also includes unsupervised models for anomaly classification and anomaly segmentation, allowing users to create pass/fail judgement models by only training with normal data. Meanwhile, smart image-labeling features allow labeling through region clicks and specific keyword inputs. The updated software also includes pre-trained models for optical character recognition (OCR).
To Learn More:
Contact: Neurocle
Headquarters: Seoul, The Republic of Korea
Product: Neuro-T and Neuro-R
Key Features: Synthetic defect data creation, Unsupervised models for anomaly classification and anomaly segmentation, Pre-trained OCR models
What Neurocle says: View more information on Neuro-T and Neuro-R.
Share your vision-related news by contacting Linda Wilson, Editor in Chief, Vision Systems Design.
