Skipping prediction in nvinfer element

[thieshanthan]

The interval property in nvinfer is useful but it has to be set befor the pipeline is initialized and it is fixed. But is there a way in the current Deepstream version to dynamically skip frames in nvinfer element depending on the output of the previous element?

• Hardware Platform (Jetson / GPU): Jetson Orin Nano
• DeepStream Version: 6.3
• JetPack Version (valid for Jetson only): 5.1.2
• TensorRT Version: 8.5.2.2
• Issue Type( questions, new requirements, bugs): Question

[Fiona.Chen]

@thieshanthan

But is there a way in the current Deepstream version to dynamically skip frames in nvinfer element depending on the output of the previous element?

What kind of skip do you need? Do you want to pass the frames to downstream even you do want them been inferenced?

[thieshanthan]

@Fiona.Chen

Do you want to pass the frames to downstream even

I need those frames to be not inferenced. i.e. I need to inference only if a condition is met from previous element.

[Fiona.Chen]

Currently there is no such function here. The gst-nvinfer is open source. You may modify it the add such function.

[thieshanthan]

thanks, is it possible to modify the interval property of nvinfer dynamically without stopping/pausing the pipeline

[Fiona.Chen]

@thieshanthan

is it possible to modify the interval property of nvinfer dynamically without stopping/pausing the pipeline

gst-nvinfer is open source. You can modify as you like. Please make sure you are familiar with GStreamer and DeepStream before you start to customize your own functions.