AVerAI DeepStream Debug Toolkit

Table of contents

• DeepStream Debug Toolkit
  • Table of contents
  • Introduction
  • Requirement
  • Quick Start
  • Notification
  • Future Works

Introduction

DeepStream Debug Toolkit can help developers to send a single image file to the deepstream pipeline and get input and output tensors from AI models with only needs python code.

The tool is suitable for developers who want to make sure the numerical correctness as expected values for a AI model running on deepstream pipeline before deploy the model into a large application.

This convenient debug tool is created by AVerMedia.

Requirement

• AVerMedia Box PC
• Docker image : DeepStream 6.3

Quick Start

Clone the tool repository.

git clone https://github.com/AVerMedia-Technologies-Inc/DeepstreamDebugToolkit.git

Build the docker image for the toolkit on AVerMedia Box PC.

cd DeepstreamDebugToolkit
docker build -t ds_kit .

Start up a container for DeepstreamDebugToolkit

docker run -it --runtime=nvidia --ipc=host --privileged -v /usr/sbin:/usr/sbin ds_kit

Run the demo.py.

python3 demo.py

Notification

• The tool can only run on AVerMedia Box PC.
• The tool can not get correct tensor arrays without mapping /usr/sbin to docker container.

Future Works

• The tool support PGIE only, not yet for SGIE. We are working on support both PGIE and SGIE.