Google Photos uses clientele to train its image recognition capabilities

Google Photos has rolled out a new survey feature in which it asks users to evaluate and describe the contents of photos as it looks to upgrade its image recognition algorithms.

According to Android Police, the current move is a follow-up to what the service did more than a year ago, by asking simple questions to customers aimed at improving the app’s facial recognition accuracy and to be able to associate photos with the right profile.

The report explains that with the current survey, the Google Photos app is asking users to label their photos from square one, rather than just confirming pre-set suggestions. It adds that the goal of this is to achieve better image and object recognition.

Android Police states further that the new option is still rolling out server-side and is selected from little card at the bottom of the Search tab. Once a user clicks on the option and agrees to respond to the survey, the app will make images and a text box pop up. The user will then be asked to type what they think is important in each particular photo.

Users can write as many descriptions of the photo as they can and then submit them. There are ten photos in the first group, but more photos can be requested depending on the user’s ability to describe them.

At the end of the process, the app asks if users are interested in other questions and exercises. One of the exercises checks if certain photos are print-worthy, the other finds out which types of animations and collages and other automated photo creations they are interested in, and the last seeks to know whether photos taken on a given day are related to that day’s known holiday, reports Android Police.

In July a database used in training systems for tasks like facial biometrics and object recognition was taken down by the Massachusetts Institute of Technology (MIT) after it was reported that it included racist, misogynistic and vulgar labels.

Article Topics

accuracy  |  algorithms  |  biometric dataset  |  biometrics  |  facial recognition  |  Google  |  image recognition  |  object recognition