ONNX is a community project created by Facebook and Microsoft.ONNX stands for Open Neural Network eXchange format. The first version was released in December 2017
The goal of ONNX is to enable Neural Network sharing between different deep learning frameworks: framework interoperability.
That’s a good idea if you don’t want to spend your time trying to convert models from one framework to another, or train from scratch a model you can’t find in the flavour you like.
Their website is straightforward and should quickly get you to using this standard on your project with their tutorials and the github.
All frameworks are not treated the same.
Framework interoperability is achieved for:
- Caffe2 (Facebook)
- Cognitive Toolkit (Microsoft)
- mxnet (Apache Foundation)
- pyTorch (Facebook now)
Some converters are available for:
- CoreML (Apple)
- Tensorflow (Google)
And runtimes are available for :
- TensorRT (Nvidia)
More info: https://github.com/onnx/tutorials
Let’s hope a full support for Tensorflow and CoreML will be achieved soon.