ML Classification

This is a library for classifying images in Unity Engine using NatML. The following ML models are included:

• EfficientNet Lite 4
• MobileNet v2
• ShuffleNet v2
• SqueezeNet 1.1

Setup Instructions

This package requires NatML, so make sure NatML is imported into your project. Then in your project's Packages/manifest.json file, add the following:

{
  "dependencies": {
    "com.natsuite.ml.classification": "git+https://github.com/natsuite/ML-Classification"
  }
}

Classifying an Image

First, assign one of the included models (in the ML folder) to an MLModelData field in your script:

using NatSuite.ML;
using NatSuite.ML.Vision;

public class Classifier : MonoBehaviour {

    public MLModelData modelData; // Assign this in the Inspector
}

Then create a classification predictor:

void Start () {
    var model = modelData.Deserialize();
    var predictor = new MLClassificationPredictor(model, modelData.labels);
}

Finally, classify an image:

Texture2D image = ...;
var (label, confidence) = predictor.Predict(image);

You cal also classify a WebCamTexture or create an image feature directly from pixel data.

Requirements

• Unity 2019.2+
• NatML 1.0+

Supported Platforms

• Android 7.0 Nougat or newer (API level 24+)
• iOS 13+
• macOS 10.15+
• Windows 10 64-bit

Quick Tips

• See the NatML documentation.
• Join the NatSuite community on Discord.
• See NatML on Unity Forums.
• Contact us at hi@natsuite.io.