On Wednesday, META released an artificial intelligence model capable of identifying specific objects within an image. Additionally, Meta published a dataset of image annotations, which it claims is the largest of its kind.
According to a blog post by the company’s research division, their Segment Anything Model (SAM) is capable of identifying objects in images and videos, even in instances where it has not been trained on those specific items.
SAM allows for object selection through either clicking or text prompts. During a demonstration, the tool successfully identified multiple cats in a photo when prompted by the written word “cat.”
Large technology companies have been promoting their advancements in artificial intelligence since the emergence of Microsoft-supported OpenAI’s ChatGP chatbot, which became popular in the autumn and prompted a surge of investments and competition to dominate the field. Meta has hinted at various features that use the generative AI technique popularized by ChatGP, which generates entirely new content instead of merely identifying or classifying data like other AI, but has not yet launched a product.
Two examples of these tools are one that creates surreal videos based on text prompts and another that generates illustrations for children’s books using written descriptions.