model2c

About the Project

Python API and Command Line tool to convert ML models into low-level inference for embedded platforms

Getting Started

Prerequisites

Make sure you have tensorflow tf2onnx or torch installed.

Furthermore, make sure you have onnx2c installed and added to PATH.

Lastly you need ProtocolBuffers libraries installed, e.g.:

• Ubuntu: apt install libprotobuf-dev protobuf-compiler
• MacOS: brew install protobuf

Get the sources:

git clone https://github.com/kraiskil/onnx2c.git
cd onnx2c
git submodule update --init

Then run a standard CMake build

mkdir build
cd build
cmake ..
make onnx2c

And finally add to path

export PATH=$PATH:/path/to/onnx2c/folder

Installation

You can can install the package through pypi:

pip install model2c

Or you can clone the repo and build directly from source:

git clone git@github.com:h3x4g0ns/model2c.git
cd model2c
make install

Usage

Train a model with correponding data until sufficient metrics are obtained.

import torch
import model2c.pytorch import convert

# run convertor
convert(model=torch_model, 
        input_shape=(batch_size, 1, 224, 224),
        quantization="fp32",
        output_file="model.c")
print(f"size of output model: {os.path.getsize('model.c')/1024} kilobytes")

Support

model2c currently supports the following ML frameworks

• torch
• tf/keras

To Do

• torch convert
• tf convert
• make command line utility
• include dynamic axis for batch size