Skip to content
This repository has been archived by the owner on Jul 10, 2023. It is now read-only.

Latest commit

 

History

History
77 lines (60 loc) · 5.45 KB

building_pipeline_server.md

File metadata and controls

77 lines (60 loc) · 5.45 KB
Raw

Building Intel(R) DL Streamer Pipeline Server

| Build Stages | Default Build Commands and Image Names | Selecting Pipelines and Models at Build Time | Supported Base Images |

The Intel(R) Deep Learning Streamer (Intel(R) DL Streamer) Pipeline Server docker image is designed to be customized to support different base images, models, pipelines, and application requirements. The following sections give an overview of the way the image is built as well as common customization patterns.

Note: Descriptions and instructions below assume a working knowledge of docker commands and features. For more information see docker documentation.

Build Stages

The Pipeline Server docker images are built in stages. Each stage can be customized to meet an application's requirements.

Stage Description
Media Analytics Base Image The Media Analytics Base Image contains a media framework plus all of its dependencies(GStreamer* or FFmpeg* ).
Intel(R) DL Streamer Pipeline Server Library Python modules enabling the construction and control of media analytics pipelines.
Models and Pipelines Deep learning models in OpenVINO™ IR format. Media analytics pipeline definitions in JSON.
Application / Microservice                                                            Application or microservice using Intel(R) DL Streamer Pipeline Server python modules to execute media analytics pipelines. By default a Tornado based RESTful microservice is included.

Default Build Commands and Image Names

Command Media Analytics Base Image Image Name Description
./docker/build.sh intel/dlstreamer:2022.2.0-ubuntu20-gpu815 docker image dlstreamer-pipeline-server-gstreamer Intel(R) DL Streamer based microservice with default pipeline definitions and deep learning models.
./docker/build.sh --framework ffmpeg --open-model-zoo... openvisualcloud/xeone3-ubuntu1804-analytics-ffmpeg:20.10 docker image dlstreamer-pipeline-server-ffmpeg FFmpeg Video Analytics based microservice with default pipeline definitions and deep learning models.

Building with OpenVINO™, Ubuntu 20.04 and Intel(R) DL Streamer Support

Example:

./docker/build.sh --framework gstreamer

Building with Ubuntu 18.04, OpenVINO™ and FFmpeg Support

Example:

./docker/build.sh --framework ffmpeg \
  --open-model-zoo-image openvino/ubuntu18_data_dev \
  --open-model-zoo-version 2021.1

Selecting Pipelines and Models at Build Time

By default the Pipeline Server build scripts include a set of sample pipelines and models for object detection, classification, tracking and audio event detection. Developers can select a different set of pipelines and models by specifying their location at build time through the --pipelines and --models flags.

Note: Selected pipeline definitions must match the media framework supported in the media analytics base image.

Specifying Pipelines and Models

Note: Pipelines(--pipelines) must be within build context.

Example:

./docker/build.sh --framework gstreamer --pipelines /path/to/my-pipelines --models /path/to/my-models

The Pipeline Server includes by default the models listed in models.list.yml in the models folder. These models are downloaded and converted to IR format during the build using the model download tool. The above example shows a directory being passed as argument to --models option. When its a directory name, the models are expected to be there. You can also pass a yml file as input with a list of models you wish to be included from Open Model Zoo.

Example:

./docker/build.sh --framework gstreamer --pipelines /path/to/my-pipelines --models /path/to/my-models.list.yml

Supported Base Images

All validation is done in docker environment. Host built (aka "bare metal") configurations are not supported. You may customize and rebuild base images from source to meet your runtime requirements.

Base Image Framework OpenVINO™ Version Link Default
Intel DL Streamer™ 2022.2.0-ubuntu20-gpu815 GStreamer 2022.2.0 Docker Hub Y
Open Visual Cloud 20.10 xeone3-ubuntu1804-analytics-ffmpeg FFmpeg 2021.1 Docker Hub Y
Intel DL Streamer™ 2022.1.0-ubuntu20 GStreamer 2022.1.0 Docker Hub N

* Other names and brands may be claimed as the property of others.