EdgeNeuron

TensorFlow Lite Micro Library in Arduino Style

EdgeNeuron is an Arduino-friendly wrapper around TensorFlow Lite Micro, simplifying the process of running Tiny Machine Learning (TinyML) models on Arduino boards. It provides a beginner-friendly, Arduino-style API, eliminating the need for complex C++ constructs like raw pointers, making it accessible to developers familiar with standard Arduino programming.

---

Features

• Intuitive API: Simplified functions for deploying TensorFlow Lite Micro models on Arduino-compatible boards.
• No Raw Pointers: Designed to avoid complex C++ constructs, enabling easy integration into Arduino sketches.
• Optimized for TinyML: Tailored for experimentation and development on boards with sufficient computational power.

---

Installation

Via Arduino IDE:

1. Open the Arduino IDE.
2. Navigate to Sketch > Include Library > Manage Libraries.
3. Search for EdgeNeuron and click Install.

Manual Installation:

1. Download the latest .zip file from this GitHub repository.
2. In the Arduino IDE, navigate to Sketch > Include Library > Add .ZIP Library.
3. Select the downloaded .zip file to install the library.

---

Usage Examples

The provided examples demonstrate how to:

• Deploy pre-trained models for inference.
• Read data from sensors, preprocess it, and input it into the model.
• Retrieve and process model outputs.

---

TinyML Development Workflow

1. Data Collection:
   Use an Arduino sketch to collect sensor data required for training your model.

2. Model Creation:
   Define and develop a deep neural network (DNN) in a TensorFlow environment (e.g., Google Colaboratory).

3. Model Training:
   Train the model using the collected dataset in TensorFlow.

4. Model Conversion:
   Convert the trained model to TensorFlow Lite format and export it as a .h file containing a static byte array.

5. Inference Sketch Preparation:
   Write an Arduino sketch to deploy the trained model for real-time inference.

6. Include Required Files:
   Add EdgeNeuron.h and your model's header file (model.h) to your sketch.

7. Tensor Arena Definition:
   Allocate memory for model operations by declaring a byte array (tensor arena).

8. Model Initialization:
   Use the modelInit() function to load the model and prepare input/output tensors.

9. Input Data Assignment:
   Populate the input tensor with data using modelSetInput().

10. Run Inference:
    Perform inference using modelRunInference().

11. Retrieve Outputs:
    Extract the results using modelGetOutput().

---

API Reference

Model Initialization

bool modelInit(const unsigned char* model, byte* tensorArena, int tensorArenaSize)

Initializes the TensorFlow Lite Micro environment, sets up the model, resolves operations, and allocates memory for tensors.
Returns:

• true if successful.
• false if an error occurs.

---

Input Data Assignment

bool modelSetInput(float inputValue, int index)

Sets the value at a specific index in the input tensor.
Parameters:

• inputValue: The input value to be set.
• index: The index within the input tensor.
  Returns:
• true if successful.
• false if the index is invalid or an error occurs.

---

Model Inference

bool modelRunInference()

Executes the inference operation using TensorFlow Lite Micro interpreter.
Returns:

• true if successful.
• false if inference fails.

---

Output Retrieval

float modelGetOutput(int index)

Retrieves the output value at a specified index in the output tensor.
Parameters:

• index: The index within the output tensor.
  Returns:
• The output value if successful.
• -1 if the index is invalid or an error occurs.

---

Future Enhancements

• Support for advanced machine learning models (e.g., object detection).
• Extended compatibility with additional TensorFlow Lite features.
• Improved optimization for memory-constrained devices.

---

Contributing

We welcome contributions to enhance EdgeNeuron!

• Submit bug reports or feature requests via the GitHub Issues page.
• Fork the repository, make changes, and submit a pull request.

---

License

This project is licensed under the MIT License. See the LICENSE file for more details.

---

Support and Documentation

Visit Consentium IoT Documentation for tutorials, examples, and additional resources.
For further assistance, email us at [email protected].

---

Badges