Unity image recognition.

Mostly suitable for handwritten sketches - when you need to recognize what user has drawn.
Uses Keras for machine learning to train model for your specific dataset, and Barracuda to run model in Unity.
Also has a tool created inside unity that allows you to create dataset fast and easy.

How to use

Dataset preparation

To prepare your dataset you will need to run unity project.
Open DatasetCreator scene, select DatasetCreator object and input folder name.

Run the scene and just start drawing on white surface - whenever you hold - draw.
Each time you release mouse - it saves the image to specified folder.

Resulted images will be save in Assets/You_folder_name

Training model

Everything related to training is stored in ML folder

Copy folders with images into ML/Data

Open ML/Final.ipynb

Run the notebook from top to bottom.

The model will be saved as Model.onnx, and class names associated with each id will be saved as ClassNames.json

Usage

Create Model Definition scriptable object.

Provide model and class names.

Open scene Drawing Guesser and provide your Model Definition scriptable object.

Run the scene and try to draw!

Code usage example you can find in DrawingGuesser.cs

Install as UPM Package for Unity

You can isntall image recognition package for Unity to easily integrate only necessary part for your project:

https://github.com/roflseech/unity-ml-test.git?path=/Assets/ImageRecognition