Welcome to follow my public account [Jizhi Vision], reply 001 to get Google programming specification

O_o >_< O_o O_o ~_~ O_o

This article describes how to use and configure DeepStream-app.

For details on how to install DeepStream, please refer to my previous articles on Installing DeepStream6.0 for Ubuntu and DeepStream5.1 for Ubuntu. /usr/bin generates some routine executables after a successful installation, such as deepstream-app, which is the most basic one.

1. Deepstream Architecture

The DeepStream architecture is shown below:

Deepstream is built with reference to GStreamer and is made up of plugins such as VIDEO DECODE, TensorRT inference engine, OpenCV tracker, display rendering plugin, etc. Here’s a list of the main plug-ins:

  • Gst-nvstreammux: this is a stream pooling plug-in that aggregates multiple input streams into a number of buffers;
  • Gst- nvDSPreProcess: Preprocessing on pre-set ROIs;
  • Gst-nvinfer: Tensorrt-based inference plug-in, which is the most core;
  • Gst- NvTracker: OpenCV based tracking plugin for target ID tracking;
  • Gst- NVMultiStreamtiler: used to form 2d array form video frames;
  • Gst-nvdsosd: This is a screen display plugin for drawing boxes, checkboxes, etc.
  • Gst-nvmsgconv and GST-NVMSGBroker: Used to transfer analysis data to the cloud server.

2. Deepstream-app configuration description

Deepstream-app is a routine executable program in the DeepStream SDK. It is also very powerful and has many configurations.

The deepStream-app configuration is broken down into blocks, including the following:

There will be a lot of configuration parameters in each specific Group, among which the plug-in Primary GIE and Secondary GIE Group is used to load TensorRT ENG for reasoning, which is clearly stated in the official NVIDIA document. Can undertake the corresponding query, there is not much said, teleport: docs.nvidia.com/metropolis/…

Take a look at some things in config:

You can see things like application, tiled-display, source, sink… All of these can be found in the query table provided by NVIDIA. If you need to modify them, you can modify them accordingly. This is also the original intention of DeepStream. Users do not need a deep understanding of how each widget is developed.

3. Use of Deepstream-app

To demonstrate a routine, do as follows:

cd /opt/nvidia/deepstream/deepstream/samples/configs/deepstream-app

deepstream -c ./source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8.txt
Copy the code

A demo video pops up, and here’s a screenshot, and it’s pretty good.


Deepstream-app can be used to configure deepstream-app. It can be used to configure deepstream-app.

Introduction to Deepstream-app