In a document released on Friday, Apple detailed the company’s latest developments in device learning technology, such as the introduction of a bust-based character recognition function for iOS 15, even if the face is not clearly recorded in the photo. At the same time, the company’s special page of machine learning research detailed how to better recognize people in photos locally on the device.
The article pointed out that the photo album application can better identify people in various scenarios, regardless of whether their faces have been clearly recorded. To achieve this, Apple uses person-specific face and bust recognition technology.
The company wrote: If the subject’s sight is moved away from the lens, the face is usually blocked or completely invisible. To this end, the development team considered the upper body of the character in the image, because it usually shows constant characteristics in a specific environment, such as unchanging clothing.
These features make it possible for iOS mobile devices to efficiently distinguish people in photos taken in a few minutes before and after. As shown in the picture above, Apple takes a complete image as input, and then specifically recognizes the face and upper body features in it.
Then, by matching these features with other photos, traditional facial recognition cannot effectively distinguish the ability of individuals. As always, Apple emphasized that all device learning operations can be done locally to protect users’ privacy.
At the same time, the company also hopes to reduce the overhead of memory and computing resources as much as possible. Interested friends can move to the Apple Machine Learning Research Portal (Portal) to view the full text of Recognizing People in Photos Through Private On-Device Machine Learning.
