Amazon’s Rekognition Gets New ML Capability for Object Recognition

“Amazon claims that its system can be used to train an AI to detect an object using as few as 1- images, as opposed to the hundreds or thousands that are usually required for object recognition.”

Amazon has announced a new machine learning capability for object recognition through its Rekognition platform.

Called “Amazon Rekognition Custom Labels”, the feature is distinguished from other machine learning systems by its capability of learning to recognize objects with a relatively limited dataset. Amazon claims that its system can be used to train an AI to detect an object using as few as 1- images, as opposed to the hundreds or thousands that are usually required for object recognition.

What’s more, Rekognition’s Senior Product Manager, Anushri Mainthia, claimed in a blog post that the feature “requires no ML experience” and the use of “only a few lines of code”.

It’s a bold claim, and one that may warrant some skepticism given Amazon’s track record with respect to Rekognition’s facial recognition capability. After calling attention to Amazon’s sale of Rekognition to police agencies, the American Civil Rights Union claimed that when it used Amazon’s system to compare images of members of Congress to a database of 25,000 mugshots, it returned 28 matches, with 39 percent of the matches being people of color despite the fact that only 20 percent of members of Congress were non-white. This kind of inaccuracy – and particularly its racial skew – suggests inadequate training of Rekognition’s machine learning technology, and the use of an image database that is too small and not sufficiently diverse.

That having been said, Amazon has disputed the claims of the ACLU and other academics who have criticized its AI technology.

The company says that NFL Media and VidMob, a marketing analytics company, are already using its Custom Labels system. Others will gain access with its wide launch on December 3rd.

Sources: TechCrunch, VentureBeat, AWS Machine Learning Blog

–

December 3, 2019 – by Alex Perala