Skip to content

Latest commit

 

History

History
93 lines (69 loc) · 6.83 KB

README_MCU.md

File metadata and controls

93 lines (69 loc) · 6.83 KB

Get Started with a microcontroller (MCU)

In the next window the project name, the output path and the neural network to be converted are passed.

In the next window you can select the target you want to execute the TensorFlow model. You can choose if you want to execute it on an MCU, an FPGA or a single-board computer (SBC). Here you select the MCUs button.

In the fourth window, the optimization algorithms pruning and quantization can be selected. It is important to know that only for fully connected and convolutional layers the pruning algorithm can be applied.

If you select pruning, you can choose between two options:

  • Factor: For the fully connected and convolutional layers, a factor is specified in each case, which indicates the percentage of neurons or filters to be deleted from the layer.

  • Accuracy: The minimum accuracy of the neural network or the loss of accuracy that may result from pruning can be specified here.

If quantization is selected, you can choose between two options:

  • int8 + float32: This quantization approach converts all weights to int8 values. But the input and output still remain 32-bit float.
  • int8 only: All weights get converted to int8 values. Also the input and output will be converted to 8-bit integer. When executing the net later, the input values of the model must be passed as signed int8 values. Also the output values are returned as signed int8 values.

The next window appears only if at least one optimization algorithm has been selected. In this window the training data are selected which the neural network requires for the optimization. The data can be transferred in different ways:

  • Path: images are to be used as training data. In the given path there are subfolders containing the name of the different classes of the neural network and the corresponding images.
  • File (CSV file): The data is stored in a CSV file. The rows of the CSV file contain the different data samples. The first or last column of the file have to contain the target label. The other columns contain the data of each sample.
  • File (Python file): The data is loaded and returned in a Python script. Here it is important that the Python script contains the function get_data() with the return values x_train, y_train, x_test, y_test (training data, training label, test data, test label). The return values here are NumPy arrays. An example of how such a file can look like can be found here.

If a CSV file is selected the CSV dataloader window opens. With the browse button a new CSV file can be selected again. By using the different separators, it is possible to define how the data is separated. The Preview button shows an overview of how the data will look with the selected settings. In addition, the label of each data series must be specified. Here it is possible to set the label to the first or the last position of a data series. In addition, the number of rows and columns in the data set is displayed. If all settings are correct, the settings can be taken over for the later optimization via the button Load data.

At the end, you have to specify the amount of memory for the input, output, and intermediate arrays of the neural network on the MCU. In addition, an overview of all selected parameters is displayed here. The button in the lower right corner starts the process of the conversion.

Model execution

The .h and .tflite models are stored in the created project folder. However, these are not required for execution on the MCU. In the folders src and inc the model is stored, as well as a script for initializing and executing the model (respectively .cpp and .h file). In addition, the folder tensorflow_library contains the files needed from the TensorFlow library to run the model on an MCU. These are inserted into the project (for Arduino projects the TensorFlow library is not needed, because it can be directly included from the Arduino IDE). The following shows some code snippets that need to be inserted into the MCU project:

// include the script to load and execute the model
#include "tf_lite_exe.h"
...
// Depending if your model output are float or int values
// define an array to store the output of the model.
float* prediction = new float[output_neurons]; 
int8_t* prediction = new int8_t[output_neurons];
// If you just have one output neuron it looks like this:
float prediction 
int8_t prediction
...
// Call the function to setup the model from tf_lite_exe 
setup_model();
...
// Call the function to execute the model. As parameter you
// pass the input data. The prediction of the model is stored
// in the prediction variable.
prediction = model_execute(input_data);

In addition, an Arduino sample project for the MNIST dataset can be found here. As already mentioned, the .cpp and .h files (mnist_data, tf_lite_exe) are simply inserted here. The only code that needs to be written can be found in the mnist.ino file. First the tf_lite_exe.h file is included. Afterwards a variable is declared to store the predictions and the functions to initialize and run the model are called.