IBM & Gado Images to Demo AI Model Release Assignment Prototype at DMLA 24th Annual Conference

October 19, 2019


Share on facebook
Share on twitter
Share on linkedin
Share on email

Visual AI can do some incredible things. In 2019, using off-the-shelf AI and Machine Learning solutions from multiple vendors, media licensors can automatically tag massive image libraries, find landmarks or logos in images, and even get full-text captions that read like they were written by a person. Optical Character Recognition (OCR) can pull text out of an image–even if it’s handwritten, or on some tiny visual element, like a road sign – and automatically translate it into 100+ languages.

These off-the-shelf capabilities are remarkable, given where AI was just a few years ago. And for many users, pre-trained models and off-the-shelf solutions provide more than enough intelligence for their application, often at very little cost. But what if your solution is more complex, and specific to the media licensing industry?

The assignment of model release forms to images is one particularly challenging problem faced by the licensing industry today. For an image to be used for commercial purposes (such as in an advertisement or political campaign), any recognizable people in the image must have signed a release form authorizing the use of their likeness. When a contributor uploads a new image to a media licensor, the licensor needs to determine whether the image contains people, and whether those people have signed release forms.

Finding people in an image might seem like a straightforward AI task at first. But look at any major image library, and the complexities are immediately clear. What if a person has their back to the camera? What if they’re wearing a mask, or using some kind of equipment which partially obscures their face, like a respirator or surgical mask? What if the image uses a creative technique like motion blur or selective focus, and some faces are blurred or distorted? All these factors make identifying people in an image, and thus assigning proper release forms, a more challenging AI problem than it first appears.

Given the complexity of assigning model releases–and the potential risk of liability if you get it wrong – many media licensing agencies pay human reviewers to look at each image they receive, manually identify people, and ensure each person who needs a release form has one on file. This is time consuming and expensive.

That’s where custom AI solutions come in. Many vendors offer custom AI solutions, but one prominent player in the industry is IBM Watson. Watson has been in the national spotlight since its pivotal 2011 win against human opponents in the TV show Jeopardy. Since that time, IBM has taken Watson from a purpose-built AI for one specific task to a platform which is revolutionising the implementation of AI and Machine Learning across a variety of industries, from manufacturing and construction to cultural heritage and media licensing.

At the DMLA’s 24th Annual Conference in October, as part of the Practical AI session featuring Google Cloud, Imagga and CloudSight, Gado Images and IBM Watson will present a prototype of a new, custom AI system designed specifically for assigning model release forms to images. The system will look at an image, automatically find all the people in the image, whether they’re looking at the camera (like in a straightforward portrait), standing in a crowd, facing away, or even wearing something that obscures their face. This feature alone is a big step towards automating many agencies’ workflows.

The system will eventually go beyond this, determining if faces are sufficiently small, obscured or selectively-focused that a release form is not needed. If the system is successful, it will offer a great example of the power of custom AI to solve specific, challenging industry problems.

The Digital Media Licensing Association’s DMLA 24th Annual Conference will take place October 27-29th in Marina del Rey, California and the Practical AI session will take place Monday morning October 28th at 11am. If you’d like to see initial results of the model release assignment system, along with new insights from Google Cloud, Imagga and CloudSight – Join us, register and see complete program at –