Teachable Machine Lite

By: Meqdad Darwish

Description

Run your Google Teachable Machine image models on Raspberry Pi, robots and other edge devices in a few lines of Python. Lightweight, powered by LiteRT, no TensorFlow required.

Teachable Machine Lite lets you take an image model you trained in Google's Teachable Machine, with no coding needed, and run it on a Raspberry Pi, a robot, or any small computer with just a few lines of Python. It runs on LiteRT (formerly TensorFlow Lite) instead of the full TensorFlow, so it installs quickly, starts fast and works comfortably on resource-constrained hardware. Pass it a camera frame from OpenCV or Picamera2 and get back the predicted class and its confidence, ready to draw on the image or act on. That makes it a good fit for robotics, IoT and other edge AI projects. It's the lightweight sibling of the Teachable Machine package.

Source Code is published on GitHub

Read more about the project (requirements, installation, examples and more) in the Documentation Website

Supported Classifiers

Image Classification: Use an exported TensorFlow Lite model from Teachable Machine Platform (a model file with tflite extension). Both the Quantized and the Floating point exports are supported.

Requirements

For detailed information about package requirements and supported platforms, please visit our documentation

Python >= 3.9

Works with NumPy 1.x and 2.x. On Raspberry Pi, use the 64-bit Raspberry Pi OS.

How to install Teachable Machine Lite Package

pip install teachable-machine-lite

On Raspberry Pi OS, install inside a virtual environment; see the installation guide.

Dependencies

numpy
Pillow
ai-edge-litert   # LiteRT (TensorFlow Lite); tflite-runtime on 32-bit Raspberry Pi OS

Example

An example for teachable machine lite package with OpenCV:

from teachable_machine_lite import TeachableMachineLite
import cv2 as cv

cap = cv.VideoCapture(0)

model_path = "model.tflite"
labels_path = "labels.txt"

tm_model = TeachableMachineLite(model_path=model_path, labels_file_path=labels_path)

while True:
    ret, frame = cap.read()
    if not ret:
        break

    results, result_image = tm_model.classify_and_show(frame)
    print("results:", results)

    cv.imshow("Camera", result_image)
    k = cv.waitKey(1)
    if k == 27:  # Press ESC to close the camera view
        break

cap.release()
cv.destroyAllWindows()

Values of results are assigned based on the content of labels.txt file.

For more, including the Raspberry Pi camera module (Picamera2), take a look at these examples

Links:

• Documentation

• Changelog

• PyPI

• Source Code

• Teachable Machine Platform