zoom_in Click to enlarge

SAM

▲ ▼

description SAM Overview

SAM (Segment Anything Model) is a vision foundation model developed by Meta AI and released in 2023. It was trained on the SA-1B dataset of over one billion segmentation masks on eleven million images, enabling zero-shot generalization to segment objects not present in training data. Users can prompt SAM with points, bounding boxes, or text to produce segmentation masks for arbitrary objects in images. The model and dataset were publicly released to support research in computer vision.

help SAM FAQ

What is the SAM model in AI?

SAM stands for the Segment Anything Model, a vision foundation model developed by Meta AI. It was released in 2023 to perform object segmentation tasks. The model can identify and cut out individual objects within an image, even objects it has never encountered before.

Who created the Segment Anything Model?

The Segment Anything Model (SAM) was created and released by the Meta AI research team. It was trained on the massive SA-1B dataset. This dataset contains over one billion segmentation masks on eleven million images.

What is the SA-1B dataset?

The SA-1B dataset is a massive collection of images and segmentation masks created specifically to train Meta's Segment Anything Model. It features over one billion masks on eleven million licensed images. It is one of the largest and most detailed segmentation datasets ever published.

How does SAM perform zero-shot generalization?

The Segment Anything Model enables zero-shot generalization, meaning it can segment objects not present in its training data. Users can provide a prompt, such as a click, a box, or text, to guide the model to a specific object. This flexibility allows it to be used immediately on new visual tasks without requiring additional training.

swap_horiz

Looking for SAM alternatives? Compare top competitors ranked & scored

arrow_forward