This is a realtime implementation of Meta AI's Segment Anything Model in Unity Engine using NatML.
The SAM projects consists of two models:
- A large image embedding model
- A lightweight mask decoder network
We setup the embedding model using a NatML endpoint (see embedding.ipynb). We export the lightweight mask decoder network to ONNX, then use NatML to convert it to CoreML for iOS; ONNX for Windows and the web; and TensorFlow Lite for Android.
- NatML 1.1.4+
- Android 24+
- iOS 14+
- macOS 10.15+
- Windows 10+ (64-bit only)
- Web:
- Chrome 91+
- Firefox 90+
- Safari 16.4+