TinyML 🍰

A convolutional neural network running on an OpenMV Cam H7 Plus board for the course in Embedded Systems.

Specifications

• Hardware: OpenMV Cam H7 Plus
• Dataset: Food-101
• Libraries: Keras, TensorFlow
• Environment: Google Colab, OpenMV IDE
• Language: Python

Description

This Convolutional Neural Network (CNN) has been trained to classify images from 3 different classes of food:

• pizza
• spaghetti carbonara
• tiramisù

and it reaches an accuracy of over 80%.

Usage

Both the cases explained below require to move the model food101.tflite inside the board's microSD.

Images from memory

Inside the folder test there are script and images to test the algorithm without the need to capture real-time images.

NOTE: images must be uploaded on the microSD!

Real-time images

If you want to use the sensor as shown in the video, it is sufficient to run the other script.

NOTE: latency in the video is higher than the expected one because this video was recorded while using an older version of the model!