An Icon Classifier with TensorFlow Lite Model Maker
AI for Art and Design
Written by Margaret Maynard-Reid, ML GDE
This is part 1 of the Icon Classifier tutorial, in which I discuss how to make an icon classifier with the TensorFlow Lite Model Maker. In Part 2, I will discuss how to implement the model on Android (with TensorFlow Lite metadata, Codegen and ML Model Binding in Android Studio).
A recent new feature in TensorFlow Lite is the Model Maker that helps you to create a model easily. All you need is to supply the data - with just a few lines of code, an image or text classifier is created without much in-depth knowledge in machine learning.
In this tutorial let’s take a look step by step how to use the TFLite Model Maker to train an icon classifier. For details please see source code in Colab notebook on GitHub here.
Step 0. Installation
Create a new Colab notebook and pip install the TFLite Model Maker in the notebook:
!pip install git+git://github.com/tensorflow/examples.git#egg=tensorflow-examples[model_maker]
Step 1. Data preprocessing
There are a few scenarios of getting the data for model training:
- Create your own.