We all have a huge number of photos on our computers these days. But how often do you find yourself searching endlessly for that photo you took three years ago on Thanksgiving because you forgot to tag it with any relevant information when it was saved? Well, Google has once again solved all your problems. They've revealed a new captioning system that can teach itself how to accurately describe a photo.
By using their deep learning methodology (basically, a set of algorithms that are used), Google is able to recognize multiple objects in a photo. For example, in this photo of a bowl of fruit, the technology is able to accurately recognize each of the fruits separately. The technology is sort of similar to Facebook's technology that takes already tagged photos and automatically tags faces that it recognizes in new photos, but this is much more advanced than that. It's not fully developed yet of course, but what they have so far is impressive.
Google has developed a new technology that can accurately describe the contents of photos.
Currently, what they are working on is further developing a natural language to help create more in depth captions that can describe precisely what is going on in said photo. So instead of the caption, "orange, orange, banana, orange, orange," it could potentially be, "fruit bowl full of bananas and oranges." Google is also attempting to teach this system how to describe more complex photos as well.
While again, this technology is not yet perfect, these advances in technology are a huge step forward. Pretty impressive if you ask me.
[via TechCrunch and Google]
The potential for this kind of scene recognition is really fun to imagine...
Can't wait to have a lightroom plug-in of this technology !!