> My pipeline is nvarguscamerasrc(capsfilter)->nvstreammux->nvinfer->nvvideoconvert->nvdsosd->nvvideoconvert->x264enc->flvmux->rtmpsink (there are queues between each element), the delay is very serious(playing through VLC, about 6s), so I replaced x264enc with nvv4l2h264enc->h264parse, but the delay did not decrease, only the CPU utilization was reduced.Could you give me some suggestions and guidance, thank you.
1.This does not mean that the delay is caused by the performance of Deepstream/Jetson nano. You can try to remove plugins such as nvstreamux/nvinfer and only use nvarguscamerasrc/nvv4l2h264enc/rtmpsink to build a pipeline to observe the latency.
You can also use nveglglessink to observe the latency
2.This FAQ can measure the element latency of each deepstream
I just replaced osd_sink_pad_buffer_probe with latency_measurement_buf_prob. Also, I set num_sources to a constant 1. This also makes me very confused.
This measurement approach determines the input timestamp based on the name of the nvv4l2decoder element, and you are using test1, which may cause some problems.