Skip to main content

Training a Custom Classifier Model

Google's Teachable Machine is a browser-based tool that allows users to create classifier models without any code or advanced understanding of machine learning.

Data Collection​

Gather images for each class you want to recognize. Ensure that you have a balanced dataset (a roughly equal number of images for each class). Ensure you have diversity in camera angle, lighting, backgrounds, object characteristics, etc.

Training with Teachable Machine​

Go to the Teachable Machine website: teachable machine

  • Click on the “Get Started” button.
  • Select the “Image Project” to create an image classifier.
  • Select "Standard image model"
  • For each class:
    • Click on the 'Add Class' button.
    • Upload the images pertaining to that class.
  • After adding all classes, click on the “Train Model” button.
  • Once training is complete, you can test the model right in the browser.
  • Click the "Export Model" Button
  • Navigate to the "Tensorflow Lite" tab
  • Select "EdgeTPU" if you have a Limelight Google Coral
  • Select "Quantized" if you are using a Limelight 3A for FTC or you are using CPU Neural Network functionality.
  • Download the model.