You are here: Home » Artificial Intelligence » Getty’s AI Tool Gets Nearer the Goal

Getty’s AI Tool Gets Nearer the Goal

by Charles Russell on September 5, 2018

Many people working within a modern Digital Asset Management environment will be aware of the new wave of AI-assisted image recognition components that are being touted as the next big thing.  An increasing number of vendors are jumping on the bandwagon, claiming that their automated solutions can liberate users from the drudgery of sourcing and tagging images.  However, as my co-contributor Ralph Windsor has previously written about in some detail, such AI-based solutions often yield less than satisfactory results, and the truth is that the technology still has some way to go before it can be compared to the decision-making skills of a human equivalent.

Stock photo agency Getty Images have recently announced a new AI-assisted tool that takes a slightly different approach to the problem of sourcing images.  ‘Panels by Getty Images’ attempts to ‘read’ an article – much as a human editor would – and subsequently sources images from its library based on the article’s content.  Additional filters and a self-improving algorithm endow the tool with the ability to progressively ‘learn’ how a content editor selects images over time, thus providing more relevant results.

The project, developed in partnership with image optimization platform Vizual.AI, is currently only available to Getty Images premium subscribers, so we can’t validate any claims or testify to its effectiveness in the field.  Senior VP of Data and Insights Andrew Hamilton says:

“In today’s digital world, publishers are under constant pressure to tell the latest story and compete for consumer attention. At the same time, we know how important compelling imagery is to creating online engagement. Panels by Getty Images meets both of these challenges for our customers using the power of artificial intelligence.”  [Read More]

The press release is accompanied by a short video, explaining how Panels analyses the copy-pasted text and pulls out keywords, weighing each word for importance and dividing them into three ‘threads’.  The operator can then add, remove and edit keywords to bring up different images from Getty’s library of over 100 million items.  Related events and themes are also offered up, which can be clicked on to provide additional threads.

Hamilton continues:

When you think about a creative process of selecting an image to match the story, it’s a much harder problem to try to solve with computers.  So even with this tool, it’s the human picture editor that has the last word. It’s not something technology can do a good enough job on.” [Read More]

Flipping the AI concept upside down to examine text rather than images strikes me as a far more valid approach.  In this scenario, the body of text that’s fed into the Panels tool has already been prepared by the editor and effectively provides the context – the one thing that’s generally missing from AI-based image recognition solutions. It’s an incredibly simple idea, and one that’s not without prior art, yet with the combined resources and vast image library that Getty has at its disposal, it’s one that looks to out-perform the current deluge of half-baked AI offerings.

Related posts:


Leave a Comment

Previous post:

Next post: