Openvx编程模型简介

An Introduction to OpenVX Programming Model

Openvx编程模型是一种新兴的计算机视觉编程模型,适合部署于对性能要求比较高的场景中,方便硬件厂商进行优化。Openvx编程模型以计算图作为核心,计算图中的节点代表一个计算操作或一个中间结果。驱动程序按照计算图中节点之间数据流动的逻辑关系执行计算图中节点代表的计算任务。执行完一个计算图中所有节点的计算任务,就完成了一次计算机视觉计算任务。驱动程序以适合硬件的方式对计算图进行调度和优化。由于深度学习模型由各种神经网络层组成,层与层之间存在数据的流动,也可以看成是一种计算图,因此openvx特别适合用于深度学习模型在边缘侧的部署和优化。本文介绍了openvx编程模型的基本概念、流程,并说明了将openvx用于部署深度学习模型的方法。本文将帮助openvx研究人员熟悉openvx,有助于对openvx的优化算法进行更深入的研究。

Openvx is an emerging programming model in computer vision field. Openvx is designed for circumstances where performance is of primary importance, and is suitable for hardware vendor to optimize. Graph is the central concept in openvx. Nodes in graph represents a computing operation or an intermediate data object. Openvx framework executes computing tasks represented by nodes in openvx graph according to edge relations between nodes. After all nodes in a graph is processed, a computer vision task is finished. Oepnvx framework can optimize graph by scheduling operations or optimizing operations according to hardware implementation. Deep learning model is comprised of layers which are linked with neuron activation flow, so it can also be regarded as a kind of graph. Openvx is particularly suitable for implementing deep learning model in embedded device. Basic concepts and processes of openvx are introduced, and mechanism of openvx to implement deep learning model is introduced. This paper will help researchers understand openvx programming model, and help to do more intensive research on graph optimization algorithms.