Meta has developed an AI model called “Segment Anything” to detect objects in pictures and videos without relying on pre-existing training data.
The AI can reliably highlight all essential elements in a photo by clicking on or typing a free-form text prompt, such as “cat.”
Related Articles
The technology can be used with other models to create 3D visuals or mixed-reality views. This new approach can decrease additional AI training.
The AI model and dataset are available for download under a non-commercial license to expand access to the technology for research purposes.
Meta uses similar technology for content moderation, post recommendations, and photo tagging.
Developers say the model needs more details and precise boundary identification than other models.
It can handle real-time prompts but not complex image processing. So, specialized AI tools may outperform it in their sectors.
Meta is known for sharing its AI breakthroughs, including a translator for unwritten languages.
Nonetheless, the company must compete with tech giants like Google and Microsoft in AI.
It is already developing generative AI “personas” for its social apps, and innovations like “Segment Anything” highlight its unique advantages.
This AI model may not be suitable for robots or other devices where fast and accurate object detection is typically required.
Yet, it may be effective in cases where training data could be more practical, such as a social network with a fast-expanding content volume.
Ultimately, this demonstrates Meta’s goal to generalize computer vision.