Clarification about process of the pipeline

[noir201]

i have the below questions which i need clarification about please

1. if we have a pipeline that uses PGIE and SGIE in deepstream plugins. Does these Nvinfers process sequentially or in parallel. I have a face detection/recognition model, can the gpu process both the detection of current frame and recognition of previous frame simultaneously or one plugin can run at a time? and what about if DLA is enabled to run detection, can gpu during this time work on the recognition of the previous frame?
2. Is it possible to enable both DLAs in Jetson Orin to run the detection model?
3. For multi input streams, does the nvstremmux handle each plugin independently or it does it process all these input as a single frame. as i below, frame count changes based on number of faces from the 4 inputs
   
   

n

[Fiona.Chen]

@noir201

I have a face detection/recognition model, can the gpu process both the detection of current frame and recognition of previous frame simultaneously or one plugin can run at a time?

The PGIE and SGIE work simultaneously on different frames.

@noir201

and what about if DLA is enabled to run detection, can gpu during this time work on the recognition of the previous frame?

Surely it can.

@noir201

Is it possible to enable both DLAs in Jetson Orin to run the detection model?

No. One DLA core for one model.

@noir201

For multi input streams, does the nvstremmux handle each plugin independently or it does it process all these input as a single frame.

nvstreammux s a plugin. As described in Gst-nvstreammux — DeepStream documentation 6.4 documentation "The Gst-nvstreammux plugin forms a batch of frames from multiple input sources. ". After nvstreammux, the downstream elements handle the batch instead of single frame in the pipeline.

@noir201

as i below, frame count changes based on number of faces from the 4 inputs

What do you mean?