Once king of AI, Meta has been experiencing the rise of generative artificial intelligence for several months. A technology that eludes him – until now at least.
However, on Wednesday, the tech giant unveiled research on image segmentation that could have a big impact on the future of AI.
Project “Segment Anything”
Image segmentation simply refers to an AI model that can identify the different elements of a photo. For example, in a photo of a crate of fruit, an AI model could, through image segmentation, identify each photographed fruit and the box, as shown in the Meta demo.
Screenshot of the SAM demo of Meta / Meta.
The “Segment Anything” project, which boils down to a new task, a new dataset, and a new model for image segmentation, aims to “democratize segmentation,” according to Meta.
Image segmentation can be used for photo editing, scientific image analysis, AR and VR
Meta has released its general Segment Anything model (SAM – Segment Anything Model) and its largest segmentation dataset, Segment Anything 1-Billion mask dataset (SA-1B). This dataset includes more than a billion masks across 11 million images, licensed and privacy-conscious, according to the group.
The social media giant says it released (SAM) and (SA-1B) “to enable a wide range of applications and promote research into basic models for computer vision.”
Image segmentation can be used for photo editing, scientific image analysis, larger artificial intelligence systems that require general multimodal understanding of the world, and most interestingly, augmented reality and reality Virtual.
Practitioners no longer have to collect their own segmentation data
Previous segmentation models either required a person to guide them through interactive segmentation or training based on substantial amounts of manually annotated objects for automatic segmentation.
SAM is a single model that can easily apply either segmentation method. This means practitioners no longer have to collect their own segmentation data and no longer need to fine-tune a model for their use case, saving them time and effort.
You can test the technology by visiting the Segment Anything demo site and uploading your own image, or using a photo from the gallery.
Source: ZDNet.com