Building a DL model for scratch for handwrittten digit recognition using MNIST dataset
Files description
my-final-model-training-and-saving.ipynb
- Building the model
- training it and obtaining the following metrics: accuracy= 99.440 | loss= 0.0216
- saving the trained model as a .h5 file
my-final-model-loading-evaluating-converting-to-c.ipynb
- importing teb previously saved model and evaluating its performance on the test data
- making a prediction of a manually loaded handwritten digit for verification purposes
- converting the keras model into a tflite model, to finally convert it to a C header file (the model was next embedded on an STM32 microcontroller)