16. Live Inference 🐱 🐶

16.1. Pre-Reading

• Picamera2 Library Docs Chapter 1. Introduction

• github.com/raspberrypi/picamera2/examples/tensorflow/real_time.py

16.1.1. Objectives

• Close the loop on the ML workflow with cats and dogs

• Conduct inference on real-life critters!

16.2. The lab

We will have a live cat and dog for you to test your system on!

Recall that the machine learning workflow is:

1. Decide on a goal

2. Collect a dataset

   • Examine and understand the data

3. Design a model architecture

   • Consider the data input pipeline

4. Train the model

5. Convert the model

6. Run inference

7. Evaluate and troubleshoot

You have accomplished most of this in a sterile testing environment. Now it’s time to use a physical input pipeline and add in real-world chaos!

16.2.1. Starting point

Complete the following code and get it to your Pi 4.

# A script that uses TensorFlow Lite and Picamera
# to classify a Cat vs. Dog.
#
# Designed to use a pre-converted fine-tuned MobilenetV2 model.