Home Artificial Intelligence Meta Unveils Image Segmentation AI Model ‘SAM’

Meta Unveils Image Segmentation AI Model ‘SAM’

0
Meta Unveils Image Segmentation AI Model ‘SAM’

Image Segmentation (Photo=Meta)

On the fifth (local time), Meta announced on its blog that it developed a latest artificial intelligence (AI) model ‘Segment Anything Model (SAM)’ that may detect objects in photos and videos, improving the unknown segmentation technology.

SAM is a picture segmentation model wherein AI separates specific objects inside a picture based on text prompts entered or points clicked. It means that you can discover and isolate specific objects inside a picture.

Until now, so as to routinely implement a segmentation model, many objects needed to be manually annotated, and interactive segmentation or learning needed to be performed based on this.

Image Segmentation Model 'SAM' (Photo=Meta)
Image Segmentation Model ‘SAM’ (Photo=Meta)

SAM, however, doesn’t must collect its own segmentation data and doesn’t should fine-tune the model to suit the use case. It might probably be utilized in various fields comparable to photo editing, scientific image evaluation, multimodal-based large AI system, augmented reality (AR), and virtual reality (VR).

Along with SAM, Meta will construct a dataset ‘SA-1B’ containing 11 million images and 1.1 billion segmentation masks and release it for research purposes.

“We’re releasing it to support a wide selection of applications with the SAM and SA-1B and to advertise further research on computer vision fundamental models,” Meta said.

In the future, SAMs could be used to identify everyday items that could alert and guide users through AR glasses.  (Photo = Meta)
In the longer term, SAMs may very well be used to discover on a regular basis items that might alert and guide users through AR glasses. (Photo = Meta)

Currently, the code for the dataset is open-sourced on GitHub, and the image segmentation technology is on the market through a free interactive demo.

On this demo, visitors can select an object after uploading a photograph to isolate the thing inside a range box, or use a feature that routinely identifies all objects throughout the image.

Meanwhile, Meta announced that it might commercialize technologies comparable to ‘ChatGPT’, a generative AI, throughout the 12 months.

Andrew Bosworth, CTO of Meta, predicted that advertisers will give you the option to provide videos with AI within the promoting business, which is the essential income for Meta, thereby significantly saving money and time. Bosworth explained that it should even be applied to Facebook, Instagram and the Metaverse.

Chan Park, cpark@aitimes.com

LEAVE A REPLY

Please enter your comment!
Please enter your name here