Segment Anything Model (SAM)

Segment Anything Model (SAM)

SAM is Meta AI Research's foundation model for image segmentation. Trained on 1 billion masks across 11 million images, it can segment any object in any image given an arbitrary prompt (point, bounding box, or text). SAM marked a watershed moment in vision AI by making segmentation as easy as prompting.

Key Features

• Zero-shot generalisation to unseen object classes without any fine-tuning
• Multiple prompt types: clicks, bounding boxes, text (with grounding models)
• Automatic mask generation for entire images without any prompts
• Three model sizes: ViT-H (huge), ViT-L (large), ViT-B (base)
• SAM 2 extends the model to video segmentation with memory mechanisms

Quick Start

pip install segment-anything
wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth

from segment_anything import sam_model_registry, SamPredictor
import numpy as np
from PIL import Image

sam = sam_model_registry["vit_h"](checkpoint="sam_vit_h_4b8939.pth")
predictor = SamPredictor(sam)

image = np.array(Image.open("image.jpg"))
predictor.set_image(image)
masks, scores, logits = predictor.predict(
    point_coords=np.array([[500, 375]]),
    point_labels=np.array([1]),
)

npx ai-supply add segment-anything-model

Curated mirror of the open-source Segment Anything Model (Apache-2.0). Get it from the source.