
Earlier this week, Facebook announced that it updated its photo recognition tools as the Facebook AI should now be better suited for searching photos, even ones without tags.
The social media platform already offered quite impressive features. Facebook’s photo recognition tools were already able to generate “auto tags”. This was able to search and index certain elements. But now the system will be even smarter.
On February 02, Joaquin Candela released an official statement. He is the Facebook Director of Applied Machine Learning. Candela shared an official blog post. In it, the company explained Facebook’s new photo recognition tools.
More exactly, its upgraded recognition skills. Previously, Facebook was able to identify specific objects in a photograph. From now on, it will also be capable of recognizing scenes. And also actions.
As such, it will be able to tell if you are sitting, walking, or running, to give just a few examples. It will also automatically distinguish between say New York and Los Angeles.
Facebook’s photo recognition tools are based on Lumos. This latter is an artificial intelligence program. Through it, a computer can analyze an image. And also “see” its various elements. Its abilities function even without a previously introduced text description.
Lumos is already being used. It is the reported base of quite a number of photo recognition tools. For example, those that can detect and flag nudity. And also its spam features.
The new upgrade, announced this week, will work as follows. It will be based on an existing software. This latter was developed so as to help the visually impaired. It can do so by describing an image.
In describing it, the software uses automatically generated alt tags. This latter has now been updated to recognize actions. For the moment, it can distinguish between 12 activities. It recognizes walking and dancing. And also verbs that can with a noun. For example, playing an instrument or riding a horse.
The program was and will still be developed for the visually impaired users. But it will also be used with the general photo recognition tools. As such, any Facebook user will be able to search for quite anything.
They will be able to find a specific place, action, or even a specific element. This is somewhat available now as well. But the update will bring a higher degree of accuracy. And it will specifically target images that lack relevant text information.
Lumos is an AI-based system. As such, it will continue learning. As new images will be introduced, it should become able to recognize new actions. The machine learning team that develops it already introduced some 130,000 photos. These were already labeled with actions. Which allowed Lumos to distinguish between them.
Users can start using the new photo recognition tools as follows. They can search for their desired element and then click the “Photos” tab. This will display relevant search results. They will be gathered from the user’s own newsfeed. And also from public and friend photos. The search will make use of their accompanying captions. But also of the AI generated tags.
According to Candela, the photo recognition tools could also come to serve the videos segment. This latter could come to benefit from the same system. Candela expressed his excitement at the technology. He considers that this is just the beginning. And that it could come to develop various new skills.
For the moment, the feature is available for United States-based users. Most probably, a complete roll out will be carried out over the following months.
Image Source: Flickr