Who is the classifier for?
The Vigil AI Classifier was originally developed for law enforcement users; it is now available to selected commercial and NGO organisations across the world.
How does the classifier work?
The classifier is built using state of the art machine learning technology; through the “training” process it learns to discriminate between different types of imagery including legal adult pornography, everyday images, and Child Sexual Exploitation (CSE) images. When deployed as a software micro-service or when integrated into a larger software application, the classifier processes images and videos and returns confidence scores for each category it has been trained on.
How is the classifier integrated and deployed?
The classifier is available as a micro-service in both Windows and Linux forms, as well as through integrations with a number of digital forensics products. It utilises GPU hardware acceleration where possible.
What types of data can the classifier process and categorise?
The classifier has been trained to recognise images according to UK CSE standards; however, its outputs can be mapped to international standards. A video-enabled version of the classifier will be available very soon.
How are the classifier’s results displayed?
This depends on how you choose to integrate and make use of the classifier. You may choose to integrate it into your existing moderation pipeline or workflow, where it is typically used to rank images in terms of severity and by category. This helps manage and streamline the moderation process as well as supporting moderators with psychological burden.
Are the classifier’s results easy to interpret?
The classifier outputs a score for each category it has been trained on; these scores are between 0 and 1. We can provide detailed guidance on how to interpret these numbers and how to ensure integration is effective.
How is the classifier licensed?
Use of the classifier is controlled by a variety of licensing mechanisms dependent on application.
What are the recommended hardware specifications for the Vigil CAID classifier?
Please contact us for this information.
What is the testing procedure for the classifier?
Testing machine learning classifiers is complex and must be approached carefully. We can provide guidance and consultation on how to perform your own independent testing of the classifier in an effective manner.
Is the software GDPR compliant?
Yes – the classifier and software have been built in accordance with GDPR.