What's difference between multi-branch and single branch in gstreamer pipeline?

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform (Jetson / GPU) :Jetson
• DeepStream Version:5.1
**• JetPack Version (valid for Jetson only)**4.5
• TensorRT Version:7.1.3

Hi guys,
Suppose I create a simple pipeline for musti-stream rtsp source like this:

Solution 1: >> one branch

Sources | streammux | nvinfer | custom_plugin | nvosd | fakesink

Solution 2: >> two branch

Sources | streammux | nvinfer | streamdemuxer | tee name=t | t. custum_plugin -name=b1| nvosd | fakesink | t. custom_plugin -name=b2 | nvosd | fakesink

The aim of custom plugin is that for capture metadata of buffer like this:

    batch_meta = pyds.gst_buffer_get_nvds_batch_meta(hash(gst_buffer))

    l_frame = batch_meta.frame_meta_list
    while l_frame is not None:
        try:
            frame_meta = pyds.NvDsFrameMeta.cast(l_frame.data)
        except StopIteration:
            break

        frame_number = frame_meta.frame_num
        l_obj = frame_meta.obj_meta_list

        while l_obj is not None:
            try:
                obj_meta = pyds.NvDsObjectMeta.cast(l_obj.data)
            except StopIteration:
                break
            try:
                l_obj = l_obj.next
            except StopIteration:
                break
            l_frame = l_frame.next
        except StopIteration:
            break

So in the solution 1 we have loop as many streams and only once do:

batch_meta = pyds.gst_buffer_get_nvds_batch_meta(hash(gst_buffer))

but in the solution 2 we had multi branches as many streams and also we get multiple batch_meta as many streams.

I want to know which of solutions can be efficient in terms of speed and resources used?

I want to know the below lines have overhead if we call multiple times in each of branches or it’s best to have one branch and only call once of them?

batch_meta = pyds.gst_buffer_get_nvds_batch_meta(hash(gst_buffer))
frame_meta = pyds.NvDsFrameMeta.cast(l_frame.data)
obj_meta = pyds.NvDsObjectMeta.cast(l_obj.data)

3

If there is streamdemuxer, tee is not needed. Gst-nvstreamdemux — DeepStream 6.3 Release documentation

Based on the correct pipeline.

The efficiency is the same.

4

Based on correct nvstreamdemux pipeline, there is no overhead.

6

@Fiona.Chen ,
In my opinion, if we create more branched that means the pipeline handles the pipeline with more threads, and this cause bad effective if the number of threads excesses from optimal limit.