Whenever you upload a photo to Facebook or one of the company's other services -- more than 2 billion each day -- tags are added to these photos.
One reason for this is that tags help visually impaired users understand photos posted on Facebook. This in turn highlights how far Facebook's image recognition algorithm has come.
Facebook may use the tagging eventually to improve existing services, think search or making sure images don't violate policies, or use it to power entirely new services.
It is also a reminder for anyone that information can be extracted from images that are not part of the image's metadata.
The algorithm that analyzes images on Facebook adds rather basic tags such as sky, tree, or outdoor to images currently.
If you are interested in how Facebook tags photos uploaded by its users to the service currently, you need to install a browser add-on that reveals that for now.
Show Facebook Computer Vision Tags is a browser add-on for
Chrome and Firefox that adds tags to photos on Facebook automatically.
Update: The Chrome extension has been removed from Google Chrome's Web Store, the Firefox add-on is still available. There is no comparable extension available for Chrome at this point in time.
All you need to do is install the extension in your browser of choice, and visit Facebook afterwards. You can check your timeline on the site for instance, and will notice that photos uploaded to the site list tags in the upper right corner now.
Please note that this is only the case for photos uploaded to Facebook, but not for images that are added to posts automatically (this happens when you post a link for instance).
The tags are rather basic currently, and you will notice that some images don't have any tags. This happens when Facebook's image recognition algorithm had issues identifying the image.
The algorithm does not get it right all the time either, even if you set aside photos that have not been tagged at all.
The algorithm added the tag "2 people" to the photo above. This is clearly wrong, but it is understandable how an algorithm could make the mistake of thinking these statues are people. Even then, it should have least identified three, if not more, people and not just two.
It does an okay job however for most photos on the site.
The takeaway from this is that Facebook has begun to label images that users upload to the site using image recognition algorithms. If you are particularly interested in how Facebook identifies images that you upload to the site, install the browser extension to find out.
If you like our content, and would like to help, please consider making a contribution: