Artificial intelligence technologies, which have been made easy and useful for everyone to use in the last year, continue to progress in a way that is difficult to follow. While GPT-4 was followed with great interest for ChatGPT, which can even interpret images, a big announcement came from Meta today.
The parent company of Facebook, Instagram and WhatsApp and the biggest supporter of the metaverse world, the company has developed artificial intelligence that can distinguish objects in images. Announced as “Segment Anything Model (SAM)”, artificial intelligence can make a big contribution especially to the augmented reality world.
What can Meta’s SAM do?
It will be more understandable to explain SAM through an example. Lift your head from the screen and take a look around. You can distinguish and describe each and every object around you individually, right? That’s exactly what SAM can do.
Of course, we have encountered technologies similar to SAM until today. Even Adobe’s subject, object, and sky selector in Photoshop was one of them. But SAM can select objects within an image with much higher accuracy. Moreover, it can do this even on images that it has not been trained before.
According to the statement shared by Meta, SAM has learned what we call an ‘object’, not the objects themselves. This allows SAM to even select objects it has never encountered before. This capability also ensures that SAM does not need extra training.
So what will SAM do for us?
According to Meta, the usage areas of SAM are very wide. This technology can offer the opportunity to export any object in the image, support the mixing of digital into our reality in the days when augmented reality becomes widespread in our lives, help scientific studies, and form the basis of new artificial intelligence technologies. The scope of use of SAM is limited only by our imagination.
How is SAM used?
Meta has created a demo page dedicated to SAM to allow everyone to try this technology. On this page, you can interact with objects on their stock images. To use SAM in your own images, you can access the GitHub page and get the source codes.
35153
