Segment Anything

Zero-shot generalization to unfamiliar objects and images without additional training

Promptable segmentation system that can segment any object in an image with a single click

Flexible integration with other systems, e.g. taking input prompts from AR/VR headset gaze or object detectors

Extensible outputs that can be used as inputs to other AI systems, e.g. object tracking, image editing, 3D reconstruction, or creative tasks

Efficient and flexible model design with a ViT-H image encoder and a lightweight transformer-based mask decoder

Trained on 11M images and 1B+ segmentation masks using a model-in-the-loop data engine