We are very pleased to announce the launch of the Machine Learning how-to guide, Add a new operator to Arm NN.
The guide provides step-by-step instructions on how to add support for a new operator to the Arm NN source code. It walks you through the entire process from adding boilerplate front-end support, to implementing the new operator as an Arm NN layer and providing backend support. Along the way, the guide will introduce and familiarize you with Arm NN concepts and coding patterns. There are chapters with detailed instructions on how to add Android NN support and parser support (Tensorflow Lite/ONNX/Tensorflow/Caffe) for the new operator. Arm NN also provides a Serializer/Deserializer utility and this guide will show you how to add support for your new operator to this too.
The list of operators supported by Arm NN is extensive but there is still plenty of scope for adding new operators. We know that this guide will be useful for developers who are interested in adding new operator support for their own personal use. However, we also hope that it will encourage developers to get involved with Arm NN and to contribute new operators to be made available for all in the Arm NN community. At the end of this guide are instructions on how to submit your new operator for review to be merged to the Arm NN master. For anyone interested in being a contributor to Arm NN, adding a new operator is a great way to get started. If your favorite model is one operator away from full Arm NN support, why wait?