Some time ago, Microsoft released a facial recognition tool that was supposed to detect you in photos. However, the algorithm wasn’t flawless, as it had difficulties in recognizing people with darker skin tones. Now, the company has revealed it has improved this tool, reducing most of its inaccuracies.
The facial recognition tool had trouble with darker skin
Microsoft’s facial recognition program, called Face API, was a bit racially biased. This means it was perfect at identifying people with whiter skin, but it faced several difficulties when it came to darker-skinned people. Now, the company has finally decided to address these problems and fix them.
This new version has a much smaller error rate when it comes to identifying people of all skin colors. The facial recognition algorithms for dark skin are now better, but the system has greatly improved its capabilities when it comes to women.
Microsoft fed the algorithm with a bigger amount of data
To achieve these improvements, Microsoft fed the algorithm a much larger amount of data. It selected special samples belonging to all skin tones, ages, and gender. Also, it focused on achieving good results regardless of all these differences, so the engineers made sure the data was as diverse as possible.
There is one specific factor that defines the quality of an AI algorithm. It’s not the code behind it, but the type of data it uses. Before, Microsoft didn’t provide it with enough data to cover the entire population. Therefore, its results when it came to darker-skinned people were poorer. The company identified the weak points of the mechanism and fed it with suitable data.
However, the new version of the facial recognition tool doesn’t only account for different skin color. Microsoft knows many other elements might be present in a photo, so the new data trained the system to recognize different hairstyles, jewelry, or other accessories. Therefore, from now on, the tool should be a lot better at spotting faces in photographs.
Image source: Pixabay