Welcome to 2026

[jangorecki1]

Hello DLStreamer Community,
Happy New Year! I'm Jan, the DLStreamer Product Manager, and I'm excited to connect with you all here in our dedicated GitHub Discussions space.

What's New for 2026

We've kicked off the year with an important update: DLStreamer now has its own dedicated repository!

Based on your valuable feedback about the difficulty of tracking updates when DLStreamer was part of the broader edge-ai-libraries repo, we've made this move to improve your developer experience. While DLStreamer remains logically part of the Edge AI Libraries ecosystem, you'll now find it much easier to follow releases, issues, and discussions specific to DLStreamer.

You can find my LinkedIn post here: LINK

What's Coming
Next week: I will share more details on our latest 2025.2 DLStreamer features and capabilities
Coming soon: Get ready for some exciting reveals! I will be sharing exclusive insights into our upcoming release candidate,

Let's Build Together
This discussion space is for you! I will be sharing regular updates, but more importantly, I want to hear from you:

• What features are you most excited about?
• What challenges are you facing?
• How can we make DLStreamer better for your use cases?

Drop a comment below to introduce yourself and let me know what you'd like to see covered in upcoming posts!
Looking forward to building an amazing 2026 together,

jan

[brmarkus]

It's surprising to see that usage of DL-Streamer is shown mainly from using gst-launch-1.0 from a console/using scripts.

Many industrial use-cases use C/C++/C# applications instead, usually implementing interactions (like modifying the pipeline, adapting parameters, interactive video-wall), but also realizing low-level or algorithmic tasks (like scene-change detection to make e.g. tracking more stable).

I would love to see (more) samples and examples using C/C++/C# as well - or at least add links to such applications; maybe calling out to the industry "where and how do you use DLStreamer in your products, we would be happy to add a link.".

[jangorecki1]

Hi Markus, @brmarkus
Thank you for this valuable feedback! You're absolutely right about the need for more diverse programming language examples beyond gst-launch-1.0.

I'm happy to share that we've already started addressing this gap. We have expanded beyond command-line examples to include Python samples, which you can find here: https://github.com/open-edge-platform/dlstreamer/tree/main/samples/gstreamer/python

Looking ahead: C/C++ examples will be considered for upcoming releases.

Speaking of practical applications, here's a great community example that showcases DLStreamer in action: https://github.com/bharath5673/DLstream

Thanks again for the feedback - it's exactly this kind of input that helps us build better resources for everyone!

jan

[brmarkus]

There are already samples, e.g. Python samples, just today a new Python sample got added.
But when the Python (only)(mainly) consists of hard-coding a pipeline like

    pipeline = Gst.parse_launch(
            f"filesrc location={args[1]} ! decodebin3 ! "
            f"gvadetect model={ov_model_path} device=GPU batch-size=4 ! queue ! "
            f"gvafpscounter ! gvawatermark ! "
            f"videoconvert ! vah264enc ! h264parse ! mp4mux ! "
            f"filesink location={output_file}"
        )

then there is not much of a difference to calling gst-launch-1.0 in the console.

[tjanczak]

Python sample apps show how to integrate user-defined logic to the pipeline (which is not possible with gst-launch method) without a need to write a custom GStreamer element:

• hello_world sample shows how to register a callback to process analytics results
• open_close_valve sample shows how to dynamically re-configure pipeline flow (stop/reconfigure/start) in response to real-time event
• prompted_detection shows how to re-configure AI model and deploy it within same application + how to create a custom sink element that one can integrate with application business logic...

Existing examples are rather trivial. Yet, the goal is not to build complex pipelines, but rather demonstrate how one can build a (python) application where DLStreamer pipeline is one element, but there is also an application-specific logic around.

Any feedback is welcome!

[brmarkus]

In the meantime the OpenVINO CPU plugin has been refined for ARM CPUs, see e.g.

• https://medium.com/openvino-toolkit/openvino-toolkit-for-arm-platforms-overview-85342887ef7d
• https://docs.openvino.ai/2025/openvino-workflow/running-inference/inference-devices-and-modes/cpu-device.html

Has someone tried DL-Streamer on ARM (e.g. RaspberryPi) systems using OpenVINO (again) (recently)?