Welcome to the wonderful world of Lobe, where you can train Machine Learning models without writing a single line of code!! Lobe is free and easy-to-use, all you have to do is take pictures and label them and Lobe does the rest.

In this guide we'll walk through training a Lobe model, exporting it, and running it on a Raspberry Pi 4.

This tutorial is part of a series which include the following tutorials:

• Tutorial 2 - Build an ML Rock Paper Scissors Game with Lobe
• Tutorial 3 - Build an ML Package Detector with Lobe

Background Knowledge

To be successful with this project, you'll need some experience with the following:

1. Setting up and using the Raspberry Pi
2. Some familiarity with using the terminal window
3. Installing the Pi Camera

First, we'll train a custom machine learning model using objects on your desk. For my model I used a pen, Lobe sticker, and a succulent plant. Feel free to use any objects you'd like!

First, download and install Lobe from the link below:

1. Open Lobe and create a new project.

2. In the Label tab, select Import (top right corner), and Camera from the drop down menu.

If this is your first time using Lobe you'll need to give it permission to use your camera.

3. In the bottom left corner, type a label for the first image.

My first label is "Pen" because it's a picture of, well, a pen!

4. Take between 10 and 20 pictures of your objects using your computer's camera.

Take pictures from different angles, in different lighting conditions, and with different hand placements to improve model accuracy. 

5. Repeat steps 3 and 4 for the rest of your objects.

Remember to add a new label for each object (e.g. "Sticker" and "Succulent").

6. Add a "Nothing" category.

This improves accuracy of the ML model. See below for more info.

Why is there always a prediction even when nothing is in the image?

Lobe will always predict one of your labels even if your image does not contain any related content. If you expect your model to see these types of images, create a ‘None’ label and add variations of these images as examples. You can use this ‘None’ label as a placeholder when waiting for relevant predictions.

In Lobe, training happens automatically as soon as you add enough images (a minimum of 5 for each label). 

When you have taken between 5 and 10 pictures of each object and training is complete, test out the model by following the steps below.

1. In Lobe, select the Use tab and choose Camera.

2. Try different placements of the objects used for training and see how the model performs.

Try other objects of the same type to check model accuracy. Also try out multiple objects at once. You will notice Lobe only makes one prediction at a time, even if there are multiple objects. See if you can trick your model into predicting the wrong thing and find ways to improve it.

3. Improve your model using the buttons in the bottom right.

My model did a great job at identifying sticky notes, but it mixed up the pen and highlighter.

Next, export your Lobe model to use on the Raspberry Pi. We'll use TensorFlow Lite which is a format that is optimized for mobile and edge devices, like the Pi. 

In Lobe, navigate to the Use tab and click Export.

Select TensorFlow Lite and select a location to save the model. We'll transfer the model to our Raspberry Pi later in the tutorial.

Now that the ML model is working on your computer, let's get your Raspberry Pi and BrainCraft HAT ready to make predictions on new images (also called inferencing). Your kit includes all the parts you need for this.

The next few steps walk you through setting up your BrainCraft HAT. We will skip audio setup since we aren't using it for this project, but if you can find the complete setup guide here.

Please note that the following steps require installing a lot of packages on the Pi and it may take a few hours to download everything.

As long as you're on the same WiFi as the Pi, you can use the IP address raspberrypi.local, as shown below:

On some newer versions of Raspberry Pi OS the Fan service fails to start. Luckily these newer version of the OS have a built in fan control you can turn on.

In command prompt or terminal connect to the pi using SSH and run the command:

sudo raspi-config

Select Performance Options

Select Fan

The default GPIO pin is 14, but the BrainCraft Hat has the fan connected to pin 4.

You can set the temperature to anything between 60-120 degrees Celsius. 80 is a good midway point in that range.

You'll need to reboot after changing the settings.

Now your fan will come on when ever the board is over 80 degrees. To check the current temperature of your board you use this command:

/opt/vc/bin/vcgencmd measure_temp

Install the Pi Camera module

Make sure you have the Pi camera module by running the following command:

Now that you have everything set up, it's time to do an initial test with the camera. This should display what the camera sees on the display.

raspistill -t 0

Exit the camera test by pressing CTRL + C

Since we're using the Pi in a headless configuration, we'll use an FTP connection to transfer files between our computer and the Pi.

Windows Instructions

Download and install WinSCP

Open WinSCP and start a New Session

Select an SFTP connection, fill in the IP address of your Pi, set the username to Pi, and put in your password.

Mac Instructions

Download and install FileZilla. When it's done installing, open the program.

Type sftp:// followed by the IP address of your Pi. Set the username to pi and put in your password.

1. Connect your Pi to a power source and wait for it boot up.

You should see a solid red light and an intermittently flashing green light.

2. Open command prompt on a PC or terminal on Mac/Linux and connect to your Pi using SSH.

Type the following command but replace the bolded number below with IP address of your Pi:

ssh pi@192.168.0.22

3. Download the GitHub folder.

Run the following commands to download the sample code from GitHub:

4. Create a new folder called model in the home directory.

Type the following commands:

5. Open the FTP connection from the previous step.

6. Copy saved_model.tflite and signature.json from your exported Lobe model to the model directory on the Pi.

7. In terminal on the Pi, run the following script to install Lobe and all it's dependences:

8. In terminal on the Pi, run the Python program lobe-basic-prediction.py

Type the following commands:

Keep testing the model on the Pi and see how it works. If you find that the prediction is consistently wrong, you can add more images to the model to improve its performance.

You can train an ML model to recognize all sorts of objects and then use the BrainCraft to trigger actions in the physical world!

To learn how to do this, check out these more advanced projects:

• Tutorial 2: Build a rock paper scissor game
• Tutorial 3: Build a package detector