Among the challenges for robots and artificial intelligence is to understand what is seen through their cameras and "senses". Therefore they need to be trained to recognize each object which takes a long time. When the object is known then how it is used can be understood. It is almost impossible to teach an artificial intelligence (AI) to recognize every object in existence one by one.
Today Meta announced the Segment Anything Model (SAM) which is their latest AI model that can identify all objects in photos and videos. When an image or video is seen by SAM, it will perform the segmentation process automatically.
Segmentation is the process of breaking down each object in an image to understand it. The interesting thing about SAM according to Meta is that it can perform segmentation on objects that it has never seen before when it was trained.
Meta provides access to SAM for researchers which also contains 11 million images used to train it and over 1.1 billion segmentations that have been produced under the Apache 2.0 license. For the public, the SAM demo can be tried for free via the link below.
Tags
APPS & GAMES