Meta has developed an AI model called “Segment Anything” that can detect objects in pictures and videos without a training set. You can click on items or type in prompts to see the AI highlight all those items in a given photo. The model can also work with other models to construct an object in 3D using an image, and limit the need for AI training.
While the model may miss small details, and sometimes takes some time to detect objects. However, it is still helpful for models to not rely solely on training data in order to keep up with rapidly growing content. The AI model and dataset will be downloadable with a non-commercial license.
Read more at Engadget.