Introduction to Barracuda
Note: The Barracuda package has been replaced by the Sentis package, which is in a closed beta phase. Refer to the Sentis documentation for more information.
You can sign up for the closed beta.
The Barracuda package is a lightweight cross-platform neural network inference library for Unity.
Barracuda can run neural networks on both the GPU and CPU. For details, see Supported platforms.
Currently Barracuda is production-ready for use with machine learning (ML) agents and number of other network architectures. When you use Barracuda in other scenarios, it is in the preview development stage.
Getting started
This documentation assumes that you are familiar with neural networks and have already trained a network that needs to be integrated into a Unity-based project.
Barracuda is a simple, developer-friendly API for neural network execution. You can start using Barracuda with as little code as this:
var model = ModelLoader.Load(filename);
var engine = WorkerFactory.CreateWorker(model, WorkerFactory.Device.GPU);
var input = new Tensor(1, 1, 1, 10);
var output = engine.Execute(input).PeekOutput();
The Barracuda neural network import pipeline is built on the ONNX (Open Neural Network Exchange) format, which lets you bring in neural network models from a variety of external frameworks, including Pytorch, TensorFlow, and Keras.
The Getting started guide takes you through a short, complete tutorial on how to run your network in Unity with Barracuda. Barracuda Starter Kit contains Barracuda usage examples.
The FAQ provides answers about the design, implementation, and usage of Barracuda.
Advanced topics
This section provides more information on the following topics:
- Supported operators: provides an overview of supported operators
- Supported platforms: provides an overview of supported platforms
- Supported architectures: provides an overview of supported architectures
- Exporting model to ONNX: explains how to export a network to ONNX
- Loading model: explains how to load a ONNX network to Barracuda
- Visualizing model: explains how to vizualize ONNX and Barracuda networks
- Using IWorker interface: explains how to run your model on different backends
- Model execution: explains how to run a model
- Model outputs: explains how to introspect the model and query outputs
- Tensors: handling data: explains how to handle data in Barracuda
- Memory management: explains how memory is managed in Barracuda
Reporting issues
If you have issues running Barracuda in your Unity project, please report them on the Barracuda GitHub repository.
Requirements
The current version of Barracuda is compatible with the following versions of the Unity Editor:
- 2019.4.29f1 and later