Skip to content

Customized Run Pipeline

Overview

When the pipeline is run, the run.sh script starts the service and performs inferencing on the selected input media. The output of running the pipeline provides the inference results for each frame based on the media source such as text, barcode, and so on, as well as the frames per second (FPS). Pipeline run provides many options in media type, system process platform type, and additional optional parameters. These options give you the opportunity to compare what system process platform is better for your need.

Start Pipeline

You can run the pipeline script, run.sh with a given pipeline profile via the environment variable PIPELINE_PROFILE, and the following additional input parameters:

  1. Media type
  2. Platform
    • core
    • dgpu.0
    • dgpu.1
    • xeon
  3. Environment Variables

Run the command based on your requirement. Once choices are selected for #1-3 above, to start the pipeline run, use the commands from the Examples section below.

Examples using different input source types

In the following examples, environment variables are used to select the desired PIPELINE_PROFILE and RENDER_MODE. This table uses run.sh to run the object_detection pipeline profile:

Input source Type Input Source Parameter Command
Simulated camera rtsp://127.0.0.1:8554/camera_X PIPELINE_PROFILE="object_detection" RENDER_MODE=1 sudo -E ./run.sh --platform core|xeon|dgpu.x --inputsrc rtsp://127.0.0.1:8554/camera_1
RealSense camera <serial_number> --realsense_enabled PIPELINE_PROFILE="object_detection" RENDER_MODE=1 sudo -E ./run.sh --platform core|xeon|dgpu.x --inputsrc --realsense_enabled
USB camera /dev/video0 PIPELINE_PROFILE="object_detection" RENDER_MODE=1 sudo -E ./run.sh --platform core|xeon|dgpu.x --inputsrc /dev/video0
Video file file:my_video_file.mp4 PIPELINE_PROFILE="object_detection" RENDER_MODE=1 sudo -E ./run.sh --platform core|xeon|dgpu.x --inputsrc file:my_video_file.mp4

Note

The value of x in dgpu.x can be 0, 1, 2, and so on depending on the number of discrete GPUs in the system.