Which SAM model size should I use for my project?

Use ViT-B (375MB) for maximum speed and edge devices, ViT-L (1.2GB) for balanced performance, and ViT-H (2.4GB) when accuracy is your primary priority.

Can SAM be used for medical or specialized imagery?

Yes, one of SAM's greatest strengths is its zero-shot transfer capability, allowing it to segment medical, satellite, and underwater images without domain-specific training.

How do I deploy SAM in a web browser?

This skill includes instructions for exporting the SAM mask decoder to ONNX format, which can then be run in-browser using ONNX Runtime Web for interactive experiences.

Does this skill support SAM 2?

This specific skill focuses on the original SAM for static images. While SAM 2 exists for video, this implementation provides the foundation for image-based segmentation tasks.

What is the Segment Anything Model (SAM)?

SAM is a foundation model from Meta AI designed for zero-shot image segmentation, capable of isolating any object in an image using simple prompts like points or boxes.

Segment Anything Model (SAM)

Name: Segment Anything Model (SAM)
Author: zechenzhangAGI

byzechenzhangAGI

•

384

数据科学与机器学习

Implements Meta AI's foundation model for high-precision zero-shot image segmentation using points, boxes, and masks.

关于

This skill integrates the Segment Anything Model (SAM) into Claude Code workflows, enabling developers and researchers to perform sophisticated image segmentation tasks without task-specific training. It provides comprehensive implementation patterns for interactive segmentation via prompt engineering, automatic mask generation, and integration with both the official SAM library and HuggingFace Transformers. Whether building annotation tools, generating synthetic training data, or processing specialized imagery like medical scans, this skill offers the technical guidance needed to deploy SAM across various model sizes and deployment environments including ONNX.

主要功能

Automatic mask generation for entire images with customizable density
384 GitHub stars
Zero-shot segmentation across any image domain without fine-tuning
Interactive prompting support using points, bounding boxes, or masks
Support for multiple model architectures including ViT-B, ViT-L, and ViT-H
Guidance for ONNX export and deployment for edge and browser environments

使用场景

Building interactive AI-assisted image annotation and labeling tools
Creating automated object extraction pipelines for e-commerce and creative design
Developing domain-specific segmentation workflows for medical or satellite imagery

关于

主要功能

Automatic mask generation for entire images with customizable density
384 GitHub stars
Zero-shot segmentation across any image domain without fine-tuning
Interactive prompting support using points, bounding boxes, or masks
Support for multiple model architectures including ViT-B, ViT-L, and ViT-H
Guidance for ONNX export and deployment for edge and browser environments

使用场景

Building interactive AI-assisted image annotation and labeling tools
Creating automated object extraction pipelines for e-commerce and creative design
Developing domain-specific segmentation workflows for medical or satellite imagery