Amazon Web Services Inc. today released a new feature for Amazon Rekognition that will enable the computer vision service to identify specific objects in images — such as products on a shelf or manufacturing components — even if they aren’t in its knowledge base.

AWS is calling the feature Custom Labels. Anushri Mainthia, the senior product manager for Rekognition, explained that companies could already use the service to classify pictured objects into broad categories, but it often wasn’t possible to identify exactly what those objects are.

Providing an example, Mainthia wrote in a blog post this morning that “an auto repair shop uses Amazon Rekognition Label detection (objects and scenes) to analyze and sort machine parts in their inventory.” In that scenario, she explained, Amazon Rekognition would by default simply label the items as “Machine Parts” without adding the name of each individual component.

Custom Labels allows companies to take it a step further. “Using Amazon Rekognition Custom Labels, the customer can train their own custom model to identify specific machine parts, such as turbocharger, torque converter, etc.,” Mainthia wrote.

Companies can train Rekognition to recognize a new object by uploading as few as 10 pre-labeled images of the item in question. Building such a granular computer vision from scratch, AWS claims, requires assembling as many as millions of training images in some cases. 

Custom Labels enables users to perform the training via a graphical interface that doesn’t require any coding. After a worker uploads their labeled images, Rekognition scans a built-in library of machine learning models to find the most suitable algorithm and automatically trains it on the photos. Users can then view the trained algorithm’s accuracy in an analytics dashboard that also provides related technical details.

A number of AWS customers have already adopted Custom Labels. NFL Media uses it to automatically label team logos, pylons and other objects in footage, so employees can more easily find snippets that contain items of interest. VidMob Inc. is employing Custom Labels to help power its namesake marketing analytics platform.

Custom Labels competes with a number of existing offerings from AWS’ rivals. Google LLC provides a service called AutoML, which likewise enables users to customize machine learning models for specific image sets without any coding, while Microsoft Corp. has Azure Custom Vision. 

Images: Amazon