Segment Anything by Meta

Segment Anything by Meta AI is an artificial intelligence (AI) model designed to enable computer vision research in order to easily segment objects in any image with a single click. The model uses a promptable segmentation system that can generalize to unfamiliar objects and images without additional training. Inputs for the segmentation system include interactive points and boxes, which can generate multiple valid masks depending on how ambiguous the prompts are. Output masks can be used as inputs for other AI systems, tracked in videos, edited for image applications or lifted into 3D for creative tasks.

The Segment Anything by Meta AI model is specifically designed to be efficient enough to power data engines. It includes an image encoder and a lightweight mask decoder that run quickly on web browsers in just a few milliseconds per prompt. For more efficient inference, the image encoder requires a GPU while the prompt encoder and mask decoder can be converted to ONNX so they run efficiently on CPU or GPU across various platforms that support ONNX runtime. Furthermore, it was trained using the SA-1B dataset which consists of over 11 million licensed and privacy-preserving images resulting in 1.1 billion segmentation masks collected.