open-edge-platform/dlstreamer

What it solves

DL Streamer simplifies the creation of hardware-accelerated video and audio analytics pipelines. It removes the complexity of writing extensive code for edge AI deployments by allowing users to build intelligence pipelines—ranging from simple object detection to complex multi-sensor systems—using simple strings or minimal Python/C++ code.

How it works

Built on the GStreamer framework and powered by OpenVINO, DL Streamer provides a set of specialized elements for inference, analytics, and output. Users can define pipelines as strings (via gst-launch-1.0) or through an API, which the framework then executes with full hardware acceleration on Intel CPUs, GPUs, and NPUs. It supports a wide variety of models from sources like Hugging Face, Ultralytics, and Geti, and integrates with messaging systems like MQTT and Kafka for result publishing.

Who it’s for

It is designed for developers building edge AI applications, specifically those targeting Intel hardware, who need to deploy video analytics, 3D sensor processing (LiDAR/Radar), or Vision-Language Models (VLMs) with high performance and minimal development effort.

Highlights

• One-line pipelines: Create working detection pipelines using a single command line.
• Hardware acceleration: Unified codebase for Intel CPU, GPU, and NPU.
• VLM & GenAI support: Ability to run Vision-Language Models like MiniCPM-V, CLIP, and Whisper within pipelines.
• Multi-sensor integration: Support for muxing/demuxing multiple RTSP streams, LiDAR frames, and radar point clouds.
• Python-first extensibility: Add custom logic via Python callbacks or elements without needing C++.
• Industry standard compliance: Supports GstAnalytics metadata for interoperability.