The company’s Segment Anything Model, or SAM, could recognize objects in pictures and videos even when it hadn’t seen those objects during training.
SAM allows users to select objects by clicking on them or by entering text commands.
However, in contrast to images, videos, and text, the segmentation data required to train such a model is not easily accessible online or anywhere else.
Meta has teased a number of features that use the generative AI made famous by ChatGPT.
Adding these generative AI “creative aids” to Meta’s products is a top focus for this year.
Meta does make use of SAM-like technologies for tasks including tagging photographs, removing objectionable information etc.
A non-commercial licence will be required to download the SAM model and dataset.