Deepstream Trilogy | Our 3 articles about NVIDIA Deepstream and how to use it

Our Open-source team, Neuralet, has focused on developing and improving various Computer Vision (CV) models for real-world applications in the past couple of years. So, we have always been looking for novel tools that help us improve our applications for practical situations. The Nvidia DeepStream SDK is one of the greatest tools we have worked with during these years. Our development team has studied and learned about this excellent Video Analytics toolkit and its Python Bindings for further customization in the past year. They employed this tool to optimize some of our CV models for real-time applications. It has been fantastic for building end-to-end video analytics services; it is fast, compatible with Nvidia GPUs, great for multistream and real-time use cases.

After all, to introduce and ease the learning process of working with DeepStream, especially its Python bindings, we started writing a series of articles about it. In these articles, we tried to introduce the essential components of the Nvidia DeepStream Pipeline. We walked through this Pipeline, explaining important Gst-Elements, Source and Sind Pads, Probes and their capabilities, Gst-Buffer, Metadata, etc. We also explained how we used DeepStream and Triton Inference Server to deploy our Adaptive Object Detection on Jetsons and X86s. Finally, in the last article, we made our hands dirty by building a Face Anonymizer as a use case to show how DeepStream Python Bindings work.

Here you can find a series of questions you’ll be able to answer by reading this trilogy.:

Thanks for sharing.
Do you have or plan anything similar using C/C++?
Unfortunately, a lot of articles and projects (including DLI) are Python-based - it’s great for learning but not very useful for production development, where C is the way to go.

Hi @dmitry.skorokhod ,
You are absolutely correct. We are small team and currently we don’t have any plan to write something for C/C++. Is this something that you are interested to collaborate? We definitely can help you to write something and we can publish it in Neuralet under your name. Let me know if you are interested to collaborate.

Well, I’m currently working on a project that uses C++ DS application running on Jetson to perform some custom video analytics on multiple camera streams.
I’m still learning DS and lack of good examples with comments is something that hinders my progress in a way (as DS in not open source), that’s why I was asking about C++ version of your work.
I checked your website and couldn’t quite get what service you are providing, could you explain?
And by the way, since yesterday can’t be reached, are you aware of that?