Deepstream-opencv-test sample app is considerably slower than gst-launch pipeline

[pramod4]

Please provide complete information as applicable to your setup.
PC: RTX 2070 Super
Deepstream: 6.2
Driver Version: 525.105.17
Docker image: deepstream:6.2-devel

I am trying to run the sample app deepstream-opencv-test. I have read the instructions and installed gst-dsexample as needed. Afterwards I build deepstream-opencv-test and everything works as expected, with the bounding boxes being blurred. However the fps is quite low. I then change the eglsink to a fake sink, and use the video file /opt/nvidia/deepstream/deepstream/samples/streams/sample_720p.mp4

  if(prop.integrated) {
    sink = gst_element_factory_make ("nv3dsink", "nvvideo-renderer");
  } else {
    //sink = gst_element_factory_make ("nveglglessink", "nvvideo-renderer");
    sink = gst_element_factory_make ("fakesink", "fake-renderer");
  }

this particular video takes about 60 seconds to complete with fake sink.

However if I create the pipeline using gst-launch with the same video and same pgie, it only takes 10 seconds to complete.

gst-launch-1.0 filesrc location=/opt/nvidia/deepstream/deepstream/samples/streams/sample_720p.mp4 ! qtdemux ! h264parse ! nvv4l2decoder ! m.sink_0 nvstreammux name=m batch-size=1 width=1280 height=720 ! nvinfer config-file-path=/opt/nvidia/deepstream/deepstream/sources/apps/sample_apps/deepstream-opencv-test/dsopencvtest_pgie_config.txt ! nvvideoconvert nvbuf-memory-type= nvbuf-mem-cuda-unified  ! 'video/x-raw(memory:NVMM), format=RGBA' ! dsexample full-frame=0 blur-objects=1 ! nvdsosd ! fakesink

Thus i’m curious how the c++ pipeline is quite slow.

[yuweiw]

Could you try to set the same resolution of nvstreammux and same memory type of nvvideoconvert and comment on the probe of the source code?

[pramod4]

@yuweiw

memory type of nvvideoconvert

After setting the same resolution, there was some improvement. However the major factor was the memory type. The gst-launch uses nvbuf-mem-cuda-unified, but the sample code uses nvbuf-mem-cuda-pinned. After changing the sample code to use nvbuf-mem-cuda-unified, the pipeline execution time is the same. Thank you very much.