An Implementation of NNEF-Darknet Neural Network Model Converter

There are various neural network engines that perform AI functions, and each neural network engine uses its own format for storing neural network graph information and weight data. To provide interoperability among various neural network engines, Khronos Group proposed Neural Network Exchange Format. In this paper, we described the NNEF-Darknet neural network model converter that enables the neural network model stored in the NNEF format to be executed in Darknet that is one of neural network engines. This paper shows the architecture and operation flow of the developed neural network model converter, and shows that this converter works normally with Yolo v3 neural network example.