OpenAI, which is heavily funded by Microsoft, caused a sensation after launching ChatGPT, setting off a wave of AI competitions. Large technology companies have touted their breakthroughs in AI technology. Meta showcased their latest progress in image segmentation and computer vision.
Meta announced an AI model called Segment Anything Model (SAM) that can recognize single objects in images and videos, even objects that have not been encountered in AI training in the past.
According to Meta’s official blog post, SAM is an image segmentation AI model that can circle specific objects in an image based on text prompts or user clicks.
Image segmentation is a computer vision (CV) process that involves dividing an image into multiple segments or regions that correspond to specific objects, in order to make the image easier to analyze or process.
Meta believes that image segmentation technology can help understand web content, develop AR applications, edit images, and automatically locate and track animals or objects in videos for academic research.
Usually, establishing an accurate image segmentation model requires highly specialized work by experts. Now through SAM, Meta hopes to reduce the requirement for professional training and knowledge in image segmentation, and promote the further development of computer vision.
In addition to SAM, Meta also creates a training data set named SA-1B (Segment Anything 1-Billion mask dataset).
This includes 11 million images licensed from a major photo company and 1.1 billion segmentation masks generated using image segmentation models.
The code for SAM is currently on GitHub, and the demo site is free to try out. Meta will make SAM and SA-1B available for research under the Apache 2.0 license.
Reuters reported that Meta CEO Mark Zuckerberg emphasized the importance of incorporating Generative AI into the company’s services this year, although no commercial products using Generative AI have been released yet.
But Meta already uses SAM-like technology on the Facebook platform for photo tagging, content moderation, and confirmation of recommended posts on Facebook and Instagram.
Although image segmentation technology is not new, SAM is able to identify objects that do not exist in the training data set and is open to academic research, which has become the biggest highlight of this technology.
The release of the SA-1B is also expected to drive a new generation of computer vision applications.