Deepstream Inferance on Specific sources


Is it possible to specify on the nvds-infer plugin config file (or other way) to what video sources apply only the inferance?

Using Deepstream 5.1

Seems no.

Gst-nvinfer — DeepStream 6.0 Release documentation (

Don’t you think is something very helpful to have?

Can you share the use case which need it? Per my understand, You only need batch those sources which need infer. You needn’t batch those source which needn’t infer.

I have two video sources, video0 and video1, and the deepstream pipeline I need to use two detection models, people detection and then face detection, for the video0 I need to detect the people, but I don’t need to inference those people to find their faces, but in the video1 source i need to find the people and then their faces, in the cropped frame.

I believe if we can have something to prevent a model to do inference in some video sources we can save resources and maximize the performance of the applications.
In my case the pipeline will be doing the inference for the two AI models in the two sources, when I only need to do it in one. I know I could make two pipelines, but I need them in the same pipeline because i’m using the tiler display.

Thanks for sharing. Let me check with internal and feedback. Thanks.

Hello any workaround to approach this yet or a plan to release this on a future version of deepstream?

Can you move to 6.0 GA - Cascaded streammux feature is available there! User should launch 2 pipelines to do inference on individual sources and output of 2 nvinfer should be fed to another nvstreammux component to combine both the outputs to send to tiler. If not moving to 6.0 nvinfer is opensource so user can implement their own code.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.