Automatic photo recognition in MacOS Sierra

Upgraded Mac Mini to MacOS Sierra over the weekend. The auto-categorization of the Photos app is a significant upgrade, as the Mini's main utility in the home is to archive photos in addition to the Internet. Archiving photos is always a time-consuming thing, especially for people who travel and love to take pictures, the number of photos is growing fast, and image recognition techniques such as face recognition can make this a lot easier.

For the character recognition, Apple can also do, my photos, most of the girls, the girl for the main part of the picture is basically recognized, but there is an exception, some of the baseball cap photos are classified separately, this is an obvious example of recognition errors, hats, should be more easily recognizable than hairstyle. On face recognition is another problem is a lot of people, photo application in this area does not have clear logic to classify this kind of photos, in fact, the characters more to a certain extent, you can simply put the photos of the approximate group together, the program should try to determine which photos are the same with some people's photos, such as the girl's large number of kindergarten photos , if there are several recognizable characters in these photos that recur, they should be sorted together.

In addition, it is the category of "Best Memories". The intent of this category should be to classify the most recently photographed photos together, the actual classification effect can be. What is the logic of finding good photos? The algorithm in judging the composition seems to be a major basis, of course, first of all to eliminate the image is not clear pictures, which is very easy. The character is too small, or the picture of the tilt of the composition seems to be excluded, so it is based on the recognition of people, to judge the composition good or bad. The obvious classification error is the classification of some documents and car photos into the "best Memories", the error of the algorithm source? Because the car is identified, and the car is generally considered an important photo? What about the I.D.? My car photo is for the insurance company, it seems to cheat the program, how does it know you are in the car or doing something else? For photos with no characters present, the program lacks the exact logic to filter the photos.

Application of automatic recognition algorithm seems to be a huge amount of computation, took a long time, the overall effect is good. Believe that this feature will be improved in the future upgrade, because the algorithm will get a lot of user feedback, after accumulating a large number of error samples, the algorithm will be optimized.

Automatic photo recognition in MacOS Sierra