This program provides the functionality to convert Keras machine learning models (and saved .h5 models) into CMSIS NN/DSP C files
- 1d CNN
- fully connected NN
- pooling
- quantization optimization
- auto detect when to quantize to q15 instead of q7
This is a port from an older project that implemented a custom floating point implementation of the 1d cnn functions. As a result, this project may have some kinks in it while I try and implement the missing functionality.