Deep Learning Streamer

GitHub Readme

System Requirements | Install Guide | Tutorial | Samples | Supported Models | Elements | Developer Guide | API Reference | Release Notes

Deep Learning Streamer (DL Streamer) is an open-source media analytics framework built on GStreamer*. It lets you build video and audio intelligence pipelines — from a simple object detection command line to a multi-stream production deployment — with minimal code, running on Intel® CPU, GPU, and NPU. DL Streamer consists of:

• Deep Learning Streamer Pipeline Framework for designing, creating, building, and running media analytics pipelines. It includes C++ and Python APIs.

• Deep Learning Streamer Pipeline Server for deploying and scaling media analytics pipelines as micro-services on one or many compute nodes. It includes REST APIs for pipelines management.

Media analytics — the analysis of video and audio streams to detect, classify, track, and count objects, events, and people — powers a wide range of real-world applications: retail store and facility analytics, warehouse and parking management, industrial inspections, safety and regulatory compliance, and security monitoring. DL Streamer gives you the building blocks to bring these solutions to production without requiring deep expertise in hardware acceleration or inference optimization — a single pipeline runs on Intel® CPU, GPU, or NPU, letting you focus on your application logic while DL Streamer takes full advantage of the available Intel® hardware.

Why DL Streamer?

Benefit	Details
One-line pipelines	Build a working detection pipeline in a single gst-launch-1.0 command
Hardware acceleration	Targets CPU, GPU, and NPU on Intel platforms from a single codebase
Cross-platform	Runs on Ubuntu 22.04/24.04 and Windows 11
VLM & GenAI ready	Run Vision-Language Models (MiniCPM-V, CLIP, Whisper) in a GStreamer pipeline
GstAnalytics compliance	Supports the GStreamer industry metadata standard for interoperability
Messaging integration	Publish inference results to MQTT or Kafka with built-in elements — no extra code
Python-first extensibility	Add custom logic as Python callbacks or full Python elements — no C++ required
30+ ready-to-run samples	Covers detection, classification, tracking, VLMs, LiDAR, radar and more
Multi-stream, multi-sensor	Mux/demux dozens of RTSP streams, LiDAR frames, and radar point clouds in a single process
Model hub support	Deploy models from Geti™ Studio, Ultralytics, Hugging Face, or any ONNX/OpenVINO IR model directly

DL Streamer uses OpenVINO™ Runtime inference back-end, optimized for Intel hardware platforms and supports over 70 NN Intel and open-source community pre-trained models, and models converted from other training frameworks. These models include object detection, object classification, human pose detection, sound classification, semantic segmentation, and other use cases: SSD, MobileNet, YOLO, Tiny YOLO, EfficientDet, ResNet, FasterRCNN, and other models.

DL Streamer incorporates GStreamer* Analytics metadata library as the primary method for presenting inference results. Find out more at discourse.gstreamer.org and in the documentation.

DL Streamer provides over 30 samples, demos and reference apps for the most common media analytics use cases. They are included in Deep Learning Streamer Pipeline Framework, Deep Learning Streamer Pipeline Server, Open Visual Cloud, and Intel® Edge Software Hub The samples demonstrate C++ and/or Python based: Action Recognition, Face Detection and Recognition, Drawing Face Attributes, Audio Event Detection, Vehicle and Pedestrian Tracking, Human Pose Estimation, Vision-Language Models (VLMs), Metadata Publishing, Smart City Traffic and Stadium Management, Intelligent Ad insertion, single- & multi-channel video analytics pipelines benchmark, and other use cases.

DL Streamer offers a long list of models and samples optimized for Intel hardware platforms, which can be used as a reference or a starting point for a wide range of applications and system configurations. These models and samples are a quick & easy way to reach high performance, then benchmark and optimize your application on your system.

Deep Learning Streamer is already used by many partners and customers leading solutions, including Open Visual Cloud Media Analytics services, NTT Software Innovation Center, Videonetics Technology Pvt. Limited, AIVID TECHVISION and others.

Testimonials

“Deep Learning Streamer (OpenVINO™) is an easy-to-use and extensible application framework, which provides a well-organized set of classes and methods. In particular, Deep Learning Streamer allows us to add user-defined post processing with gvapython elements. This feature will help us develop AI-based video analytics applications for NTT’s businesses, addressing various customer demands responsively.” — Takeharu Eda, Senior Research Engineer, NTT Software Innovation Center

“Deep Learning Streamer Pipeline Server has helped TIBCO Software to develop and optimize Project AIR solution with less effort and shorter TTM, and to deliver better user experience that includes no-code data pipelines. Project AIR was able to easier deploy and expose optimized video analytics pipelines as microservices accessible for consumption via REST APIs.” — Miguel Torres, Director of the Americas - Office of the CTO at TIBCO Software

---

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

Initial steps

New to Deep Learning Streamer? Start here:

• Install Guide — install DL Streamer on your system.

• Tutorial — learn the basics by building your first pipelines.

• Samples — explore ready-to-run sample applications.