Today, Amazon Web Services (AWS) announced Amazon Rekognition Custom Labels, a new feature of Amazon Rekognition that enables customers to build their own specialized machine learning (ML) based image analysis capabilities to detect unique objects and scenes integral to their specific use case. For example, customers using Amazon Rekognition to detect machine parts from images can now train a model with a small set of labeled images to detect “turbochargers” and “torque converters” without needing any ML expertise. Instead of having to train a model from scratch, which requires specialized machine learning expertise and millions of high-quality labeled images, customers can now use Amazon Rekognition Custom Labels to achieve state-of-the-art performance for their unique image analysis needs.

To better understand Amazon Rekognition Custom Labels, let’s walk through an example of how you can use this new feature of the service.

An auto repair shop uses Amazon Rekognition Label detection (objects and scenes) to analyze and sort machine parts in their inventory. For all these images, Amazon Rekognition successfully returns “machine parts”.

Using Amazon Rekognition Custom Labels, the customer can train their own custom model to identify specific machine parts, such as turbocharger, torque converter, etc. To start, the customer collects as few as 10 sample images for each specific machine part that they would like to identify.

Using the service console, customer can upload and label these images.

No machine learning expertise is required at this stage. Customers are guided through each step of the process within the console.

Once the dataset is ready and fully labeled, customers can put Amazon Rekognition Custom Labels to work with just one click. Amazon Rekognition automatically chooses the most effective machine learning techniques for each use case.

On completion of training, customers can access visualizations to see how each model is performing and get suggestions of how to further improve their model.

In our example, the auto repair shop can now start analyzing images to detect specific machine parts by their names, automating inventory management, by using a fully managed easy-to-use API built for large-scale image processing.

Amazon Rekognition Object and Scene detection returns “Machine Parts”, while Amazon Rekognition Custom Labels trained with a few labeled images returns “Turbocharger”, “Torque Converter”, and “Crankshaft”.

Now let’s look at how customers like the NFL and Vidmob are using Amazon Custom Labels.

• NFL Media, part of the National Football League, manages an exponentially-growing library of videos and images that is difficult to search for relevant content such team logos, pylons, or foam fingers with traditional methods. Amazon Rekognition Custom Labels makes that easier, says Brad Boim, NFL Senior Director of Post Production and Asset Management.

“By using the new feature in Amazon Rekognition, Custom Labels, we are able to automatically generate metadata tags tailored to specific use cases for our business and provide searchable facets for our content creation teams. This significantly improves the speed in which we can search for content and, more importantly, it enables us to automatically tag elements that required manual efforts before. These tools allow our production teams to leverage this data directly and provides enhanced products to our customers across all of our media platforms.”

• VidMob is a marketing creative platform that provides an end-to-end technology solution for all of a brand’s creative needs with a single integrated platform combining first-of-a-kind creative analytics with best-in-class creative production to transform marketing effectiveness. Alex Collmer, VidMob CEO says,

“With the introduction of Amazon Rekognition Custom Labels, marketers will be equipped with advanced capabilities within our Agile Creative Studio, enabling them to build and train the specific products (custom labels) that they care about within their ads, at scale, within minutes. Using VidMob’s integration of Amazon Rekognition, customers have historically been able to identify common objects but now the new ability for custom labels will make our platform even more targeted for every business. With a lift of 150% in creative performance and 30% reduction in human analyst time, this will extend their ability to measure their creative performance using VidMob’s Agile Creative Studio.”

AWS customers can now easily train high-quality custom vision models with a reasonably small set of labeled images. Doing this requires no ML experience, and with only a few lines of code customers can access Amazon Rekognition’s easy-to-use fully managed Custom Labels API that can process tens of thousands of images stored in Amazon S3 in an hour.

Amazon Rekognition Custom Labels will be generally available on December 3, 2019. Click here to be notified when the service becomes available. To learn more, visit https://aws.amazon.com/rekognition/custom-labels-features/.

About the author

Anushri Mainthia is the Senior Product Manager on the Amazon Rekognition team and product lead for Amazon Rekognition Custom Labels. Outside of work, Anushri loves to cook, explore Seattle and video-chat with her nephew.