Segment images zero-shot with the Segment Anything Model (SAM)
Use Meta AI's Segment Anything Model for zero-shot image segmentation, generating masks from point, box, or mask prompts without task-specific training.
Requirements
- Install the segment-anything package along with transformers 4.30.0 or newer and torch 1.7.0 or newer; the setup works on Linux, macOS, and Windows.
- Download a model checkpoint before use: ViT-H (2.4 GB), ViT-L (1.2 GB), or ViT-B (375 MB), each provided as a downloadable file.
How to use
Choose SAM when you need to segment any object in images without task-specific training.
It also suits building interactive annotation tools with point or box prompts.
You can use it to generate training data for other vision models.
It fits cases requiring zero-shot transfer to new image domains.
It is also useful for building object detection and segmentation pipelines.
It can process medical, satellite, or other domain-specific images.
Pick from multiple model sizes: ViT-B is the fastest, ViT-L is mid-sized, and ViT-H is the most accurate.
In some cases, use alternatives instead.
Use YOLO or Detectron2 for real-time object detection with classes.
Use Mask2Former for semantic or panoptic segmentation with categories.
Use GroundingDINO combined with SAM for text-prompted segmentation.
Use SAM 2 for video segmentation tasks.
The model sizes trade off speed and accuracy: vit_h (ViT-H) is 2.4 GB, slowest, and best in accuracy; vit_l (ViT-L) is 1.2 GB with medium speed and good accuracy; vit_b (ViT-B) is 375 MB, fastest, with good accuracy.
Install the package and any optional dependencies as shown in the source commands.
# From GitHub
pip install git+https://github.com/facebookresearch/segment-anything.git
# Optional dependencies
pip install opencv-python pycocotools matplotlib
# Or use HuggingFace transformers
pip install transformers
Download the checkpoint for the model size you chose.
# ViT-H (largest, most accurate) - 2.4GB
wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth
# ViT-L (medium) - 1.2GB
wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_l_0b3195.pth
# ViT-B (smallest, fastest) - 375MB
wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_b_01ec64.pth
Run basic segmentation with SamPredictor.
import numpy as np
from segment_anything import sam_model_registry, SamPredictor
# Load model
sam = sam_model_registry["vit_h"](checkpoint="sam_vit_h_4b8939.pth")
sam.to(device="cuda")
# Create predictor
predictor = SamPredictor(sam)
# Set image (computes embeddings once)
image = cv2.imread("image.jpg")
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
predictor.set_image(image)
# Predict with point prompts
input_point = np.array([[500, 375]]) # (x, y) coordinates
input_label = np.array([1]) # 1 = foreground, 0 = background
masks, scores, logits = predictor.predict(
point_coords=input_point,
point_labels=input_label,
multimask_output=True # Returns 3 mask options
)
# Select best mask
best_mask = masks[np.argmax(scores)]
Alternatively, use the HuggingFace Transformers integration.
import torch
from PIL import Image
from transformers import SamModel, SamProcessor
# Load model and processor
model = SamModel.from_pretrained("facebook/sam-vit-huge")
processor = SamProcessor.from_pretrained("facebook/sam-vit-huge")
model.to("cuda")
# Process image with point prompt
image = Image.open("image.jpg")
input_points = [[[450, 600]]] # Batch of points
inputs = processor(image, input_points=input_points, return_tensors="pt")
inputs = {k: v.to("cuda") for k, v in inputs.items()}
# Generate masks
with torch.no_grad():
outputs = model(**inputs)
# Post-process masks to original size
masks = processor.image_processor.post_process_masks(
outputs.pred_masks.cpu(),
inputs["original_sizes"].cpu(),
inputs["reshaped_input_sizes"].cpu()
)
Review the model architecture.
SAM Architecture:
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ Image Encoder │────▶│ Prompt Encoder │────▶│ Mask Decoder │
│ (ViT) │ │ (Points/Boxes) │ │ (Transformer) │
└─────────────────┘ └─────────────────┘ └─────────────────┘
│ │ │
Image Embeddings Prompt Embeddings Masks + IoU
(computed once) (per prompt) predictions
Segment objects using point prompts.
# Single foreground point
input_point = np.array([[500, 375]])
input_label = np.array([1])
masks, scores, logits = predictor.predict(
point_coords=input_point,
point_labels=input_label,
multimask_output=True
)
# Multiple points (foreground + background)
input_points = np.array([[500, 375], [600, 400], [450, 300]])
input_labels = np.array([1, 1, 0]) # 2 foreground, 1 background
masks, scores, logits = predictor.predict(
point_coords=input_points,
point_labels=input_labels,
multimask_output=False # Single mask when prompts are clear
)
Segment objects using box prompts.
# Bounding box [x1, y1, x2, y2]
input_box = np.array([425, 600, 700, 875])
masks, scores, logits = predictor.predict(
box=input_box,
multimask_output=False
)
Combine points and boxes for more precise control.
# Box + points for precise control
masks, scores, logits = predictor.predict(
point_coords=np.array([[500, 375]]),
point_labels=np.array([1]),
box=np.array([400, 300, 700, 600]),
multimask_output=False
)
Refine masks iteratively using previous logits as mask input.
# Initial prediction
masks, scores, logits = predictor.predict(
point_coords=np.array([[500, 375]]),
point_labels=np.array([1]),
multimask_output=True
)
# Refine with additional point using previous mask
masks, scores, logits = predictor.predict(
point_coords=np.array([[500, 375], [550, 400]]),
point_labels=np.array([1, 0]), # Add background point
mask_input=logits[np.argmax(scores)][None, :, :], # Use best mask
multimask_output=False
)
Generate all masks automatically with SamAutomaticMaskGenerator.
from segment_anything import SamAutomaticMaskGenerator
# Create generator
mask_generator = SamAutomaticMaskGenerator(sam)
# Generate all masks
masks = mask_generator.generate(image)
# Each mask contains:
# - segmentation: binary mask
# - bbox: [x, y, w, h]
# - area: pixel count
# - predicted_iou: quality score
# - stability_score: robustness score
# - point_coords: generating point
Customize automatic generation with density, quality, and region-area settings.
mask_generator = SamAutomaticMaskGenerator(
model=sam,
points_per_side=32, # Grid density (more = more masks)
pred_iou_thresh=0.88, # Quality threshold
stability_score_thresh=0.95, # Stability threshold
crop_n_layers=1, # Multi-scale crops
crop_n_points_downscale_factor=2,
min_mask_region_area=100, # Remove tiny masks
)
masks = mask_generator.generate(image)
Filter generated masks by area, predicted IoU, or stability score.
# Sort by area (largest first)
masks = sorted(masks, key=lambda x: x['area'], reverse=True)
# Filter by predicted IoU
high_quality = [m for m in masks if m['predicted_iou'] > 0.9]
# Filter by stability score
stable_masks = [m for m in masks if m['stability_score'] > 0.95]
Process multiple images by reusing the predictor.
# Process multiple images efficiently
images = [cv2.imread(f"image_{i}.jpg") for i in range(10)]
all_masks = []
for image in images:
predictor.set_image(image)
masks, _, _ = predictor.predict(
point_coords=np.array([[500, 375]]),
point_labels=np.array([1]),
multimask_output=True
)
all_masks.append(masks)
Apply multiple prompts per image with a single image encoding.
# Process multiple prompts efficiently (one image encoding)
predictor.set_image(image)
# Batch of point prompts
points = [
np.array([[100, 100]]),
np.array([[200, 200]]),
np.array([[300, 300]])
]
all_masks = []
for point in points:
masks, scores, _ = predictor.predict(
point_coords=point,
point_labels=np.array([1]),
multimask_output=True
)
all_masks.append(masks[np.argmax(scores)])
Export the model to ONNX format.
python scripts/export_onnx_model.py \
--checkpoint sam_vit_h_4b8939.pth \
--model-type vit_h \
--output sam_onnx.onnx \
--return-single-mask
Run inference with the exported ONNX model using onnxruntime.
import onnxruntime
# Load ONNX model
ort_session = onnxruntime.InferenceSession("sam_onnx.onnx")
# Run inference (image embeddings computed separately)
masks = ort_session.run(
None,
{
"image_embeddings": image_embeddings,
"point_coords": point_coords,
"point_labels": point_labels,
"mask_input": np.zeros((1, 1, 256, 256), dtype=np.float32),
"has_mask_input": np.array([0], dtype=np.float32),
"orig_im_size": np.array([h, w], dtype=np.float32)
}
)
Build an annotation-tool workflow that segments on mouse clicks.
import cv2
# Load model
predictor = SamPredictor(sam)
predictor.set_image(image)
def on_click(event, x, y, flags, param):
if event == cv2.EVENT_LBUTTONDOWN:
# Foreground point
masks, scores, _ = predictor.predict(
point_coords=np.array([[x, y]]),
point_labels=np.array([1]),
multimask_output=True
)
# Display best mask
display_mask(masks[np.argmax(scores)])
Extract objects with transparent backgrounds as RGBA output.
def extract_object(image, point):
"""Extract object at point with transparent background."""
predictor.set_image(image)
masks, scores, _ = predictor.predict(
point_coords=np.array([point]),
point_labels=np.array([1]),
multimask_output=True
)
best_mask = masks[np.argmax(scores)]
# Create RGBA output
rgba = np.zeros((image.shape[0], image.shape[1], 4), dtype=np.uint8)
rgba[:, :, :3] = image
rgba[:, :, 3] = best_mask * 255
return rgba
Segment regions of interest in medical images, converting grayscale scans to RGB.
# Process medical images (grayscale to RGB)
medical_image = cv2.imread("scan.png", cv2.IMREAD_GRAYSCALE)
rgb_image = cv2.cvtColor(medical_image, cv2.COLOR_GRAY2RGB)
predictor.set_image(rgb_image)
# Segment region of interest
masks, scores, _ = predictor.predict(
box=np.array([x1, y1, x2, y2]), # ROI bounding box
multimask_output=True
)
Inspect the mask data structure returned by the automatic mask generator.
# SamAutomaticMaskGenerator output
{
"segmentation": np.ndarray, # H×W binary mask
"bbox": [x, y, w, h], # Bounding box
"area": int, # Pixel count
"predicted_iou": float, # 0-1 quality score
"stability_score": float, # 0-1 robustness score
"crop_box": [x, y, w, h], # Generation crop region
"point_coords": [[x, y]], # Input point
}
Encode masks to and decode from COCO RLE format using pycocotools.
from pycocotools import mask as mask_utils
# Encode mask to RLE
rle = mask_utils.encode(np.asfortranarray(mask.astype(np.uint8)))
rle["counts"] = rle["counts"].decode("utf-8")
# Decode RLE to mask
decoded_mask = mask_utils.decode(rle)
Manage GPU memory by using a smaller model or clearing the CUDA cache.
# Use smaller model for limited VRAM
sam = sam_model_registry["vit_b"](checkpoint="sam_vit_b_01ec64.pth")
# Process images in batches
# Clear CUDA cache between large batches
torch.cuda.empty_cache()
Optimize speed with half precision or fewer automatic-generation points.
# Use half precision
sam = sam.half()
# Reduce points for automatic generation
mask_generator = SamAutomaticMaskGenerator(
model=sam,
points_per_side=16, # Default is 32
)
# Use ONNX for deployment
# Export with --return-single-mask for faster inference
If you hit problems: for out-of-memory errors, use the ViT-B model or reduce image size; for slow inference, use ViT-B or lower points_per_side; for poor mask quality, try different prompts or box plus points; for edge artifacts, use stability_score filtering; for missed small objects, increase points_per_side.
Links
Source: nousresearch/hermes-agent and its contributors. License: MIT.
Copyright (c) 2025 Nous Research
Translation and editorial adaptation by ClawCases.
>Resources
>Related Use Cases
Agent Merge-Conflict Arbiter
Resolve a git merge conflict between two agents' branches as an impartial third party: a neutral arbiter that takes both diffs and their stated intents and produces a merged result without overwriting either side.
Autonomous Project Management with Subagents
Coordinate projects autonomously: a main agent delegates work to PM subagents that track tasks in a shared STATE.yaml, update it as they progress, pick up unblocked work in parallel, and keep all task state in version-controlled files.
Control Philips Hue lights with the OpenHue CLI
Control Philips Hue lights and scenes through a Hue Bridge with the OpenHue CLI.