ML for IoT pipeline - vcs512

Pipeline to train and evaluate metrics of Convolutional Neural Networks (CNNs) that fit in IoT/MCU (Internet of Things / Microcontroller) devices.

ML Pipeline

Notebooks were made to follow a standard Machine Learning (ML) development to production pipeline:

• COMPLETE_INFORMATION: Complete general information concluded on each step

Automatic scripts

For automatic usage in CLI, it is possible to run scripts in scripts.

Model and general parameters

Module that gather model and general parameters used: development module

Model and metrics results

Directory to save the models and its results: model dir

Requirements installation

$ pip install -r requirements.txt

References

Implementation of a model in an ESP32 MCU:

• https://github.com/vcs512/micro-cnn

Dataset reference

• https://www.kaggle.com/datasets/ravirajsinh45/real-life-industrial-dataset-of-casting-product

General pipeline

• https://www.kaggle.com/code/ginsaputra/visual-inspection-of-casting-products-using-cnn/notebook
• https://www.kaggle.com/code/sfbruno/image-processing-cnn

TinyML

• https://github.com/tensorflow/tflite-micro/blob/main/tensorflow/lite/micro/examples/hello_world/train/train_hello_world_model.ipynb
• WARDEN, Pete; SITUNAYAKE, Daniel. TinyML. O'Reilly Media, Incorporated, 2019.