Realeyes Accelerates Video Attention and Emotion Analysis with Scalable Annotation Solution

“Our integration with Appen's platform allowed us to scale up our annotation operations by collecting multiple types of high-quality annotations for thousands of hours of video footage, all in a short amount of time."

-Or Amir, Vice President of Operations - Realeyes

The Company

Founded in 2007, Realeyes is a leader in attention measurement. By using front-facing cameras and the latest in computer vision and machine learning technologies, Realeyes measures attention and emotion of opt-in participants as they watch video content online. This empowers brands, publishers and technology platforms to inform and optimize their content as well as target the right videos to the right audiences. Realeyes’ technology applies facial coding to predictive, big-data analytics, driving bottom-line business outcomes for brands and publishers.

To accurately read people’s reactions to advertisements, Realeyes has developed and continues to update their machine learning algorithms. Their algorithms are then used to help Realeyes’ customers optimize their content to be emotion-driven and interesting to viewers so they have higher-quality, revenue-driving advertising.

 

The Challenge

While building their business, Realeyes encountered a few different challenges:

• Achieving the right speed of interpreting and understanding large amounts of data
• Accessing a culturally diverse group of people as they expanded their product internationally
• Keeping client data secure in the process at all times

One of the biggest challenges that Realeyes has had, as they grew their client base, is the ability to annotate and label data quickly to deliver their product to their customers. The process for analyzing facial expressions and labeling each data point can be tedious, especially when you have a small in-house team and a growing demand for your products.

Another challenge that Realeyes had is serving customers in many different markets around the world. Facial expressions and emotional reactions are unique to culture, meaning a smile in South America may not mean the same things as it does in Japan. To account for the nuances of different cultures, Realeyes has built a large database of high-quality training data that it’s using to teach their machine learning algorithm more subtlety in interpreting attention and emotional reactions. With time, technology will become better at recognizing the difference between a big smile with full teeth and a mild, polite smile with closed lips.

The third challenge for Realeyes is data security and making sure client, as well as annotator data is kept safe. Because they’re collecting video footage from front-facing cameras, it includes people’s faces. Part of their company mission is to ensure that the footage they collect is completely safe and no one else can watch it. Realeyes uses the Appen Data Annotation Platform API to manage the video data and keep it secure.

The Solution

With Appen, Realeyes enhanced their ability to scale by collecting, analyzing, and labeling more data efficiently, without losing quality. Appen’s human annotators, as well as the advanced quality controls in the Appen Data Annotation Platform enabled Realeyes to deliver evaluations more quickly at high quality.

Realeyes works with Appen to create customized pools of contributors to ensure high-quality data output. They leverage Appen Data Annotation Platform multi-step jobs to only send client data to qualified annotators. Realeyes also makes use of Appen’s quality assurance tools, such as asking contributors control questions to ensure they’re paying attention and accurately labeling data.

When working with diverse video footage from around the world, Realeyes had to make sure they take into account the cultural nuances as they send results back to their customers. That’s where the breadth and depth of the Appen Crowd came in. Realeyes can now pair human annotators with video footage from the same specific markets to ensure that cultural nuances are caught and they’re getting high quality data on demand.

One of the most important things that Appen has been able to provide Realeyes is customizability. Realeyes has a very specialized use case. With Appen, they’ve been able to build custom tools, which have then been integrated into Appen’s platform.

The Results

By partnering with Appen’s data annotators, Realeyes has been able to exponentially increase their data annotation speed. What used to take them three months in-house, now takes just two weeks with Appen tools and contributors.

This increase in speed has not happened at the expense of accuracy. To ensure they’re getting the highest quality data possible, Realeyes programmatically runs each piece of data by multiple annotators and checks if they are annotating data in similar ways, leveraging the built-in quality controls on the Appen Data Annotation Platform. When multiple data annotators give the same result, Realeyes knows they’re getting accurate data.

Realeyes also shared with us that one of the big reasons they continue to work with Appen is that they can integrate their tools with Appen and run projects automatically, saving them time in-house. Additionally, they share that the relationship and customer service we provide by working with unique use cases and needs keeps them coming back to Appen year after year.