Deepstream multiple network engines plugin problems

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Please provide complete information as applicable to your setup.

• Hardware Platform (Xavier)
**• DeepStream Version4.0
• JetPack Version (4.2.3)
**• TensorRT Version5.1
• NVIDIA GPU Driver Version (valid for GPU only)
problems when using multiple network engines:

  1. I’m trying to using yolo-v3 network as primary engine (car detection) and sample secondary engine (car color/mode/type detection) as follow, but not work. The deepstream-test2 sample works good and yolo-v3 also runs good. Do I have to use the same network architecture when using primary/secondary detection?
  2. Can I run multiple sources and multiple networks in deepstream? For example I have two cameras connected in, one links to a network engine and another one links to a different network engine then output together. Can I do this all in one pipeline?
3

Do I have to use the same network architecture when using primary/secondary detection?

No, besides deepstream-test2, you can also refer to deepstream back-to-back sample - https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/tree/master/back-to-back-detectors

Can I run multiple sources and multiple networks in deepstream? For example I have two cameras connected in, one links to a network engine and another one links to a different network engine then output together. Can I do this all in one pipeline?

I think we can support this, I need check what’s the best pipeline for DeepStream for this use case. Will get back to you later.

Thanks!

4

For 2nd quesiton, you can create a pipeline like following if you need to put the 2 different output together, otherwise you can use 2 different pipelines or apps.

src1                  -->nvinfer1
      -->mux-->demux              -->(custom plugin) ...
src2                  -->nvinfer2
6

hello, could you please help as to how this pipeline could be achieved and how to create custom plugin?

1 Like