- Contents
Deep Learning Streamer Pipeline Server (DL Streamer Pipeline Server) is a Python-based, interoperable containerized microservice for easy development and deployment of video analytics pipelines. It is built on top of GStreamer and Deep Learning Streamer (DL Streamer), providing video ingestion and deep learning inferencing functionalities.
Video analytics involves the conversion of video streams into valuable insights through the application of video processing, inference, and analytics operations. It finds applications in various business sectors including healthcare, retail, entertainment, and industrial domains. The algorithms utilized in video analytics are responsible for performing tasks such as object detection, classification, identification, counting, and tracking on the input video stream.
Note: Detailed documentation for Deep Learning Streamer Pipeline Server can be found on the documentation website.
Follow the steps in this section to quickly pull the latest pre-built Deep Learning Streamer Pipeline Server docker image followed by running a sample usecase.
-
Clone the repository and change to the docker directory inside DL Streamer Pipeline Server project.
git clone <link-to-repository> cd <path/to/dlstreamer-pipeline-server/docker>
-
Pull the image with the latest tag from registry
# Update DLSTREAMER_PIPELINE_SERVER_IMAGE in <path/to/dlstreamer-pipeline-server/docker/.env> if necessary docker pull "$(grep ^DLSTREAMER_PIPELINE_SERVER_IMAGE= .env | cut -d= -f2-)"
-
Run the image with the compose file provided in this repo.
docker compose up
Once the container is up, we will send a pipeline request to Deep Learning Streamer Pipeline Server to run a detection model on a warehouse video. Both the model and video are provided as default sample in the docker image.
We will send the below curl request to run the inference.
It comprises of a source file path which is warehouse.avi, a destination, with metadata directed to a json fine in /tmp/resuts.jsonl and frames streamed over RTSP with id pallet-defect-detection. Additionally, we will also provide the GETi model path that would be used for detecting defective boxes on the video file.
Open another terminal and send the following curl request
curl http://localhost:8080/pipelines/user_defined_pipelines/pallet_defect_detection -X POST -H 'Content-Type: application/json' -d '{
"source": {
"uri": "file:///home/pipeline-server/resources/videos/warehouse.avi",
"type": "uri"
},
"destination": {
"metadata": {
"type": "file",
"path": "/tmp/results.jsonl",
"format": "json-lines"
},
"frame": {
"type": "rtsp",
"path": "pallet-defect-detection"
}
},
"parameters": {
"detection-properties": {
"model": "/home/pipeline-server/resources/models/geti/pallet_defect_detection/deployment/Detection/model/model.xml",
"device": "CPU"
}
}
}'The REST request will return a pipeline instance ID, which can be used as an identifier to query later the pipeline status or stop the pipeline instance. For example, a6d67224eacc11ec9f360242c0a86003.
-
To view the metadata, open another terminal and run the following command,
tail -f /tmp/results.jsonl
-
RTSP Stream will be accessible at
rtsp://<SYSTEM_IP_ADDRESS>:8554/pallet-defect-detection. Users can view this on any media player e.g. vlc (as a network stream), ffplay etc
To check the pipeline status and stop the pipeline send the following requests,
-
view the pipeline status that you triggered in the above step.
curl --location -X GET http://localhost:8080/pipelines/status
-
stop a running pipeline instance,
curl --location -X DELETE http://localhost:8080/pipelines/{instance_id}
Now you have successfully run the Deep Learning Streamer Pipeline Server container, sent a curl request to start a pipeline within the microservice which runs the Geti based pallet defect detection model on a sample warehouse video. Then, you have also looked into the status of the pipeline to see if everything worked as expected and eventually stopped the pipeline as well.
You can build either an optimized or an extended DL Streamer Pipeline Server image (for both Ubuntu22 and Ubuntu24) based on your use case. The extended image contains the Geti SDK, the OpenVINO Model API and ROS2 on top of the optimized image.
Note: Ensure to set the right values in the .env file for building DL Streamer Pipeline Server optimized image and DL Streamer Pipeline Server extended image when you follow the below steps. The mentioned file has the necessary details written as comments.
Add the following lines in .env file if you are behind a proxy.
http_proxy= # example: http_proxy=http://proxy.example.com:891
https_proxy= # example: https_proxy=http://proxy.example.com:891
no_proxy= # example: no_proxy=localhost,127.0.0.1Update the following lines in .env file for choosing the right base and target images and also for naming the image that gets built.
# See .env file for example values
BASE_IMAGE=
# See .env file for example values
DLSTREAMER_PIPELINE_SERVER_IMAGE=
# See .env file for example values
BUILD_TARGET=Note: If you do not have access to the above mentioned BASE_IMAGE, then you can build DL Streamer docker image from source and use it as BASE_IMAGE in the above mentioned .env file
Run the following commands:
To enable GPU/NPU you must first grant the container user access to GPU/NPU device(s).Because Docker Compose does not evaluate shell expressions, you need to determine the render group ID on the host system and define/export it as an environment variable before running Docker Compose. You can add group ID in [WORKDIR]/edge-ai-libraries/microservices/dlstreamer-pipeline-server/docker/.env
To check the render ID group you can use below command:
stat -c "%g" /dev/dri/render* | head -1cd docker
source .env # sometimes this is needed as docker compose doesn't always pick up the necessary env variables
docker compose buildThe docker image of DL Streamer Pipeline Server is now built (based on the .env changes done above) and available for you to run.
docker compose upRefer to the relevant Get Started section to run default sample upon bringing up Intel® Deep Learning Streamer Pipeline Server container.
Legal compliance: Build a docker image containing sources for GPL/LGPL/AGPL binary distributed components
docker build -f sources.Dockerfile -t intel/dlstreamer-pipeline-server:2025.2.0-sources .Run the following command to view the directory structure of source code for distributed GPL/LGPL/AGPL components:
docker run intel/dlstreamer-pipeline-server:2025.2.0-sources- Understand the components, services, architecture, and data flow, in the Overview
- For more details on advanced configuration, usage of features refer to Advanced User Guide.
- For more tutorials refer here