Meta has unveiled a novel AI model called “Segment Anything” that can identify objects in images and videos even if they weren’t part of the training set.
This innovative technology allows users to select items by clicking on them or using free-form text prompts. For example, typing the word “cat” will prompt the AI to highlight all felines in a given photo.
The “Segment Anything” model can work in conjunction with other models, helping to reconstruct objects in 3D from a single image or draw views from mixed reality headsets. This effectively reduces the need for additional AI training.
Meta is offering both the AI model and a dataset for download with a non-commercial license, which means creators cannot use it for product development. The primary goal is to facilitate research and broaden access to the technology.
At the moment, Meta uses similar technology to moderate content, suggest posts, and tag photos.
Nonetheless, the developers acknowledge that the current model has some drawbacks. It might miss small details and isn’t as accurate in identifying boundaries as other models.
Although “Segment Anything” can manage real-time prompts, it struggles when processing complex images. In certain areas, more specialised AI tools might perform better than this model, as Meta points out.
While this AI may not be suited for robots or devices requiring fast, accurate object detection, it could still prove useful in situations where relying solely on training data is impractical.
For instance, a social network might use this technology to manage a rapidly increasing volume of content. This demonstrates Meta’s commitment to advancing generalised computer vision.
Meta has a history for sharing its AI advancements, such as this AI model that can learn to translate languages. But the company is under pressure to prove it can compete with tech powerhouses like Google and Microsoft in the AI domain.
By developing generative AI “personas” for social apps and creating innovations such as “Segment Anything,” Meta demonstrates its own distinct strengths.
