Generate Code for TensorFlow Lite (TFLite) Model and Deploy on Raspberry Pi

This example shows how to generate code that performs inference by using a Tensorflow™ Lite model and deploy the code on Raspberry Pi™ hardware. This example uses a pretrained TensorFlow Lite model for the image classification network Mobilenet-V1 that is available on the TensorFlow webpage for Hosted models. In example, you use the codegen command to generate a PIL MEX function that runs the generated executable on the target hardware. This workflow can be used for both int8 and float TensorFlow Lite models.

This example is supported for host Windows® and Linux® platforms.

Third-Party Prerequisites

• Raspberry Pi hardware

• TensorFlow Lite library (on the target ARM® hardware)

• Pretrained TensorFlow Lite Model

Download Model

Run this script to download the image classification network Mobilenet-V1 from the URL mentioned below.

if ~exist("model.tgz","file")
    disp('Downloading 5 MB Mobilenet-V1 model file...');
    url = "https://storage.googleapis.com/download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.5_224.tgz";
    websave("model.tgz",url);
    untar("model.tgz");
end

Downloading 5 MB Mobilenet-V1 model file...

The tflite_predict Entry-Point Function

The loadTFLiteModel function loads the Mobilenet-V1 model into a TFLiteModel object. The properties of this object contain information about the model such as the number and size of inputs and outputs of the model.

net = loadTFLiteModel('mobilenet_v1_0.5_224.tflite');
disp(net);

  TFLiteModel with properties:

            ModelName: 'mobilenet_v1_0.5_224.tflite'
            NumInputs: 1
           NumOutputs: 1
            InputSize: {[224 224 3]}
           OutputSize: {[1001 1]}
           NumThreads: 8
                 Mean: 127.5000
    StandardDeviation: 127.5000

In this example, you generate code for the entry-point function tflite_predict.m. This function loads the Mobilenet-V1 model into a persistent network object by using the loadTFLiteModel function. To optimize performance, after creating the network object, set the NumThreads property based on the number of threads available on your hardware board.

The tflite_predict function performs prediction by passing the network object to the predict function. Subsequent calls to this function reuse this persistent object.

type tflite_predict.m

function out = tflite_predict(in)
persistent net;
if isempty(net)
    net = loadTFLiteModel('mobilenet_v1_0.5_224.tflite');
    % To optimize performance, set NumThreads property based on the number 
    % of threads available on the hardware board
    net.NumThreads = 4;
end
out = net.predict(in);
end

On the Raspberry Pi hardware, set the environment variable TFLITE_PATH to the location of the TensorFlow Lite library. For more information on how to build the TensorFlow Lite library and set the environment variables, see Prerequisites for Deep Learning with TensorFlow Lite Models (Deep Learning Toolbox).

Generate PIL MEX Function

Create Code Configuration Object

To generate a PIL MEX function for a specified entry-point function, create a code configuration object for a static library and set the verification mode to 'PIL'. Set the target language to C++.

cfg = coder.config('lib', 'ecoder', true);
cfg.TargetLang = 'C++';
cfg.VerificationMode = 'PIL';

Set Up Connection with Raspberry Pi

Use the MATLAB Support Package for Raspberry Pi Hardware function raspi to create a connection to the Raspberry Pi.

In the following code, replace:

• raspiname with the name of your Raspberry Pi board

• username with your user name

• password with your password

r = raspi('raspiname','username','password');

Configure Code Generation Hardware Parameters for Raspberry Pi

Create a coder.hardware object for Raspberry Pi and attach it to the code generation configuration object.

hw = coder.hardware('Raspberry Pi');
cfg.Hardware = hw;

Copy TensorFlow Lite model to Target Hardware

Copy the TensorFlow Lite model to the Raspberry Pi board. On the hardware board, set the environment variable TFLITE_MODEL_PATH to the location of the TensorFlow Lite model. For more information on setting environment variables, see Prerequisites for Deep Learning with TensorFlow Lite Models (Deep Learning Toolbox).

In the following command, replace targetDir with the destination folder of TensorFlow Lite model on the Raspberry Pi board.

r.putFile('mobilenet_v1_0.5_224.tflite',targetDir)

Generate PIL MEX

On the host platform, run the codegen command to generate a PIL MEX function tflite_predict_pil.

codegen -config cfg tflite_predict -args ones(224,224,3,'single')

Run Generated PIL MEX

Read the input image by using imread. Resize the input to required input size of the network.

I = imread('peppers.png');
I1 = single(imresize(I,[224,224]));

Run the generated PIL MEX by passing the resized input.

predictionScores = tflite_predict_pil(I1);

Map the prediction scores onto the image.

DisplayPredsonImage(predictionScores, I);