Meta develops AI model that can detect unseen objects in pictures and videos

Meta, the social media giant, has announced a breakthrough AI model called "Segment Anything." 

This model can detect objects in pictures and videos, even if they were not part of the training set, which means that it does not require previous training data to identify new objects. With Segment Anything, you can select items by clicking them or using free-form text prompts. 

For instance, you can type "cat" and watch the AI highlight all the felines in a given photo. The developers acknowledge that the existing model is not perfect, but it is a significant step toward making computer vision more generalizable. 

The model might miss finer details and may not be as accurate at detecting boundaries as some models. Nonetheless, the AI model and a dataset will be downloadable with a non-commercial license, primarily for research purposes and expanding access to the technology. 

This breakthrough has potential implications for social media, where the volume of content is growing rapidly. This technology could help keep up with the demand for moderating content, recommending posts, and tagging photos. 

Meta is no stranger to sharing AI breakthroughs, such as a translator for unwritten languages. With inventions like Segment Anything, Meta shows that it has a few advantages of its own and plans to continue leading the way in AI innovation. 

While this AI model may not be vital for fast, accurate object detection in robots or other devices, it is an essential step in the right direction. The AI community eagerly anticipates what Meta will come up with next. Meta's goal is to generalize computer vision, and we can't wait to see what other innovations they have in store for us.

Written By

Manish Content Strategist

With a mixture of literature, cinema, and photography, Manish is mostly traveling. When he is not, he is probably writing another tech news for you!

Follow