Tensorflow Lite for Micrcontroller (TFLM) is a framework that is a subset of Tensorflow which is designed to execute machine learning model on resources constrained devices i.e. microcontrollers.
The following repository will provide anyone having a STM32F429I-DISC1 the ability of executing this TinyML application which is:
A model train with The MNIST database which could make the inference on the number drawn on the LCD. From 0 to 9.
CMSIS-NN: Efficient Neural Network Kernels for Arm Cortex-M CPUs are a collection of efficient neural network kernels developed to maximize the performance and minimize the memory footprint of neural networks on Cortex-M processor cores.
Here, it maximize the performance almost 10 times.
| Model | Time [ms] |
|---|---|
| 2020 | ~1956 |
| 2020-CMSIS-NN | ~199 |
model_running_cmsis-nn.webm
model_running.webm
The model was trained in this Colab
reymor