Significantly lower false rejection rates

Since Alex Krizhevsky and his research group at the University of Toronto published their convolutional neural network in 2012 to accurately classify images in the ImageNet dataset and demonstrated that AI-based algorithms are capable of drastically outperforming traditional rules-based computer vision solutions, the research community has not looked back. AI—more specifically, deep learning—has the capability to provide much higher accuracy rates for a variety of image analysis tasks, including classification, segmentation, and generation.

In visual inspection contexts, this translates to much higher detection rates of both known and unknown defects and lower false rejection rates—down to less than 0.1%.

Easily train your own AI models

Our solution is based on the Criterion AI Platform, which is a web-based software-as-a-service tool for collecting data (typically, images), setting up and training state-of-the-art AI models, and validating your AI-based solutions in accordance with the requirements laid out by the authorities.

The Criterion AI Platform hosts a variety of different AI models, which have been carefully designed to solve common tasks in the visual inspection of pharmaceutical products and medical devices. Most of our AI models designed to analyze images are based on a modern architecture called a convolutional neural network (CNN). CNNs excel in extracting important features from images with the purpose of either classifying them, locating objects inside of them, segmenting them, or generating new realistic examples.

Deploy AI models to your production line

After having trained your AI models and evaluated their performance in the Criterion AI Platform, you can easily export them from the platform and deploy them to your production line. We partner with NVIDIA, the world’s leading provider of hardware for training and running AI models, and we make use of their Jetson TX2 and TX2i circuit boards to run trained AI models in production. We wrap these boards into an industrial casing with useful outputs such as PoE ports to connect and power industrial cameras as well as CAN bus and GPIO ports to integrate with automation devices (such as PLCs and SCADA systems).

Compliance with relevant authorities

The Criterion AI Platform, which runs in the cloud, and Criterion AI Edge—the software running on our embedded Jetson TX2-powered devices—is developed, tested and validated against the requirements laid out by GAMP 5 and CFR Title 21 Part 11. In addition, we implement the recommendations laid out by the U.S. FDA’s CGMP Guidance on Data Integrity. This means that all of the uploaded data, trained models and created deployments in the Criterion AI Platform are controlled in accordance with the requirements laid out by the authorities.

Moreover, the Criterion AI Platform offers easy-to-use to tools for setting up and running Process Qualification (PQ) tests in the cloud, making it incredibly simple to evaluate the performance of the models trained in the platform and to provide the necessary documentation to certify their ability to accurately inspect products. You can learn more about how we support and ensure compliance on our validation page.

Interested in learning more? Please contact us so we can get started with improving your visual inspection lines.

This website collects usage data to help us improve your experience of the site. Click here to accept or here to opt out.