Recall the royal wedding of Prince Harry and Meghan Markle during May this year at the St. George’s Chapel in Windsor Castle! Those watching the wedding through Sky News’ livestream had a stunning treat. Who’s Who Live, a function powered by Amazon Web Services Rekognition and its partners, came alive to help identify the royals and other notable guests with on-screen captions, relevant information about them and their connections with the royals.
Launched with Rekognition Image during late 2016, followed by Rekognition Video in 2017, now this deep learning-based managed service comes with state-of-the-art capabilities with absolutely low-cost facilitating rapid integration.
During the re:Invent 2018 Jon Turow, Head of Product, Rekognition presented a deep-dive session on how the offering makes it easy to add image and video analysis to the applications to make it easy for industries in media content discovery and use live events for social media, influencer marketing and historical media libraries.
One just needs to load an image or a video to the Rekognition API, and this AI-based managed service easily identifies the objects, people, text, scenes, and activities, as well as detect for any inappropriate content. The new features added to this function are specifying the location of common objects, like people, dogs, cars etc. With these features Rekognition can now be used by industries for media discovery, customer engagement, industrial usages, safe content, public safety, mapping and many more such activities. “The bounding boxes specify the location of common objects such as dogs, cars, people etc. You can use bounding boxes to infer how many of each object (e.g. 3 cars) and the relationship between objects (e.g. person on a skateboard) See image below:
The proof of the pudding was two distinct use cases that were discussed along with the deep dive on Rekognition.
Tinder Makes Rekognition the Tool of Its Choice
One of the cases presented was by Tom Jacques, VP Engineering, Tinder, the location-based social discover app for dating – on how the company enhances its customer experience using Rekognition.
“Tinder ingests 40 TB+ of data every day from events that feed into our data intelligence engines. Who would you like to see? Who will match with you? Who will chat with you? What content should we allow? How can we present the content? We face many challenges answering these questions for our members in real time at scale,” says Tom.
“Rekognition allows Tinder to unlock the stories behind the photos. The managed computer vision at cloud-scale handles billions of images. It allows to have privacy by design, which means separate APIs provide control to use only what we want, like objects and scene recognition,” adds Tom.
The recent changes in Rekognition have helped Tinder provide rich data with enhanced model accuracy, hierarchical taxonomy and bounding boxes. These have provided the company with deeper knowledge of scenery. “We’re excited to incorporate these features and further unlock the power of photos everywhere on Tinder.”
News UK and the Future of Image Workflow with Rekognition
The second case presented was by Rudi De Sousa, Director of Engineering and Architecture, News UK, a British news conglomerate, and a wholly owned subsidiary of the American mass media group News Corp. The company wanted to achieve improvement in their image workflow, process the entire content of past 233 years, create new, exciting and innovative products for its readers and explore cost reduction opportunities.
“Our experiment started as a proof of concept by supplementing existing IPTC metadata with metadata supplied by Rekognition on nearly 53,000 images. Rekognition helps us detect the presence, position, orientation and key landmarks of faces in the images. Now we plan to continue using Rekognition for our workflows and prepare for the mountains of metadata in the future,” says Rudi.
With the help of Rekognition, News UK can now prioritise and suggest images based on suitability to the content. It now has significantly richer metadata that enhances its workflows considerably.
Jon Turow also mentioned about a few other use cases including one from KSTAR Group of Korea, where the company wanted to make event ticketing a little bit more fun by introducing a ‘face-based’ ticketing. “The customers who opt-in, don’t have to present a physical ticket. They can have their face matched against the image they provide and walk in. As a result of this Rekognition-based app, the lines got shorter, and the events became more fun.
While Amazon Rekognition competes with Google Vision, which is an older product in the market, it offers a very competitive pricing and features. The sentiment analysis capabilities and rotation-invariant deep-learning algorithms seem to out-perform the competition. It also comes with advanced features such as Face Comparison and Face Search, which makes it attractive for users across industries.