IBM has unleashed a databased of 100 million videos and photos to help developers wanting to integrate facial recognition into their apps and services utilise machine learning better.
The company's collection contains a million individual faces, all annotated with tags describing facial features, such as craniofacial measurements, facial symmetry, age and gender. The company hoped the data will help train third-party recognition platforms to more accurately identify key traits.
"Facial recognition technology should be fair and accurate," John Smith, a fellow and lead scientist at IBM, told CNBC. "In order for the technology to advance it needs to be built on diverse training data."
The data has been pulled from a Flickr dataset and each face has a unique combination of tags to get as wide a spread of variations as possible to help train facial recognition systems to identify a wide range of features.
"Many prominent datasets used in the field are too narrow and fall short in coverage and balance," said Smith. "The data does not reflect the faces we see in the world."
IBM's database will help eradicate bias in facial recognition technology, the company explained, which is a key challenge in the research field. Earlier this week, Amazon's Rekognition facial recognition tech was highlighted as not identifying race or gender fairly. IBM itself has also come under fire for having a 35% error rate when it came to identifying darker-skinned females.
The images collated and tagged by IBM come from the huge collection of Flickr Creative Commons images, which was developed five years ago. This means they are available and free for anyone to use, for any purpose.
