SAM2 Web Demo

This repository contains a demo project for the article:

Bringing AI to the browser: Running SAM2 for interactive image segmentation

The project demonstrates how to use the Segment Anything Model 2 (SAM2) for interactive image segmentation directly in the web browser using ONNX Runtime Web (ort). It allows users to upload images, interactively add positive or negative points, and see the segmentation mask update in real-time—all running entirely client-side.

Features

• Client-Side Execution: All computations are performed in the browser, ensuring user privacy and a responsive user experience.
• Interactive Segmentation: Users can refine segmentation results by adding positive (foreground) or negative (background) points.
• ONNX Runtime Web Integration: Leverages ort to run the SAM2 encoder and decoder models efficiently in the browser.

Article

For a detailed explanation of how this project works, please refer to the accompanying article:

Bringing AI to the browser: Running SAM2 for interactive image segmentation

License

This project is licensed under the MIT License.

Acknowledgments

• ONNX Runtime Web
• Segment Anything Model 2 (SAM2)