Skip to content

Latest commit

 

History

History
66 lines (42 loc) · 2.74 KB

README.md

File metadata and controls

66 lines (42 loc) · 2.74 KB
Raw

ONNX-mlpack Translator

ONNX-mlpack Translator

Unlock the Power of Other Frameworks in mlpack

This repository contains a converter for certain machine learning models from onnx to mlpack format. Currently this repository is still under construction and might undergo some major refactoring, please use with cautios.

The repository is in developing phase and its been tested on the following models.

Models Graph Generation Weight Transfer
mobileNet ✔️ ✔️
yolo-tiny v2 ✔️ ✔️
Iris classification ✔️ ✔️

ONNX-mlpack Repository Setup Guide:

Prerequisites

  1. MLPack Installation:

  2. ONNX Installation:

    • If you don't have Protobuf installed, ONNX will internally download and build Protobuf during its build process. You only need to build ONNX. Refer to the official ONNX build instructions for more details.
    • However, to avoid potential version issues in the future, we have provided a zipped format of ONNX in the build_onnx repository along with a script that will directly install ONNX on your system.

follow the below instruction to build onnx and make the repository running:

Steps to Build ONNX

  1. Clone the Repository:

    • Clone the onnx-mlpack repository to your local system and navigate to the repository directory.

  2. Build ONNX:

    • Run the following commands to build ONNX: chmod +x run.sh ./run.sh

    • This will generate all the necessary build files for ONNX inside the build_onnx folder.

  3. Verify mlpack and ONNX Build:

    • With both ONNX and mlpack built, it's time to test the setup with an example repository.

Running the Example Repository

  1. Go to the example/iris-classification folder.

  2. In the Makefile, update the mlpack header path to match your mlpack build path. For example: -I/home/your_username/mlpack/build/installdir/include

  3. Run the make command and check the console output to verify that everything is working correctly. If everythig goes fine you can similarly run the other example as well.