Home - UviDTE-FPSoC/Zynq7000-dnn-inference GitHub Wiki

Inference of Deep Neural Networks in FPSoC devices using high level libraries

This document intends to be a guide which shows the process to use the Xilinx Edge AI tools with a ZedBoard SoC device. Most of the documentation, tutorials and examples provided by Xilinx in order to use their AI libraries and tools have been created for boards with Zynq MPSoCs Ultrascale+ chips and others. Never the less, the hardware IP block created by Xilinx to run DNN inference on their boards, the Deep Learning Procesing Unit (DPU), supports its implementation on Zynq-7000 family chips. ZedBoard mouns a XC7Z020-1CLG484C Zynq-7000 AP SoC, which is compatible with this hardware description block. In this guide it is intended to explain the whole process that has to be followed to implement DNN inference on Zedboard. Software tools that have to be installed, creation of the hardware description project, configuration of an Operating System project to use this hardware description with ZedBoard, installation of Edge AI compilation tools and inference of several state of the art DNN models such as mobilenetv1, mobilenetv2, inceptionv1 and inceptionv3. The guide is organized as follows. First, a step-by-step guide of how to install all the software programs and tools is provided. A walkthrough on how to create a description hardware to include the DPU IP block in the ZedBoard programmable logic is included as well. Once all the tools are installed, the indications to properly configure your Embedded Operating System (PetaLinux 2019.2) to run inference in ZedBoard is given. Finally, exmaples on how to execute the DNN models on ZedBoard are created, providing a function to measure the time the device uses to execute each model.

Table of contents

  1. Prerequisites
  2. Software Installation
  3. FPSoC hardware description project
  4. PetaLinux project configuration to run Deep Neural Networks
  5. DNNDK v3.1 package workflow
  6. Inference of DNNs with the DNNDK package examples
  7. Inference of AI Model Zoo DNNs with the DNNDK package