
Meta's universal image segmentation model
SAM excels in zero-shot segmentation and generality; point/box prompts for any object. + params, million-scale data; strong generalization. Best suited for segmentation research, product matting and annotation, CV apps.
Segment Anything (SAM) is Meta's universal image segmentation model. Zero-shot segmentation for any image. Point, box prompts; open-source; for CV research and apps.
Difficulty: Intermediate
Zero-shot
Segment new images without training
Point/Box Prompts
Point or box to specify region
General
Any image, any object
Image Matting
Product, portrait matting
Annotation
Dataset and semi-auto annotation
Open free; HuggingFace, ModelScope.
Zero-shot; no training for new images; point/box prompts.
Real reviews and feedback from users