How to draw marker/box in the ourput video of Deepstream YoloV3 project?

I want to draw some marker in output video of the Deepstream ObjectDetector_Yolo sample
I original run deepstream-app -c deepstream_app_config_yoloV3_tiny.txt by 2 rtsp sources
I modify the actions in nvdsinfer_custom_impl_Yolo/nvdsparsebbox_Yolo.cpp/static std::vectordecodeYoloV3Tensor(…
But I found that I cant draw any marker from nvdsparsebbox_Yolo.cpp/static std::vectordecodeYoloV3Tensor(… ?
It seems that I need to recall the frame of the Yolo output?
Does anyone know how to do that?
Marker something in the output of Deepstream ObjectDetector_Yolo sample?

Can you clarify what you mean by marker in output ? Do you want to draw some text/shapes ?
If yes, you should update the osd metadata from the a probe add the required shapes/text.

What changes did you do in the - nvdsparsebbox_Yolo.cpp/static std::vectordecodeYoloV3Tensor(… ?

Thanks a lot, I do want to draw some shapes in output
What I want to do is to distinguish something (for example, people) by deepstream sdk&yolo, and then control some iot device.
I want to do four actions in the - nvdsparsebbox_Yolo.cpp/static std::vectordecodeYoloV3Tensor(…

  1. distinguish more than two rtsp sources?
  2. judge the yoloV3 detector objects in which position of the screen
  3. control some iot device by modbus
  4. draw some shapes in the output realtime video by deepstream-app -c deepstream_app_config_yoloV3_tiny.txt
    Actions 2 & 3 I had done in the nvdsparsebbox_Yolo.cpp/static std::vectordecodeYoloV3Tensor(…
    But actions 1 & 4 I wonder know is it suitable to did chanege in the nvdsparsebbox_Yolo.cpp/static std::vectordecodeYoloV3Tensor(…?
    Or could you tell me another suggestions?

And How to update the osd metadata from the a probe add the required shapes/text.?
Could you provide me some reference documents?
How to update the osd metadata? is it suitable to did change in the nvdsparsebbox_Yolo.cpp/static std::vectordecodeYoloV3Tensor(…?
Or I need to do change in another documents? or another func?
Thanks for you kindly reply again, If you know the answer of my questions, Please give me some suggestions, Thanks

Best regards,


Hi @cycheng0122,
is your question addresed in How to judge different rtsp sources of objectDetector_Yolo multi rtsp resources sample? ?


it may not proper to add metadata extraction in
nvdsparsebbox_Yolo.cpp/static std::vectordecodeYoloV3Tensor(…
you may refer to code, for how to draw text by adding probe on element, draw other shapes also support,
then do some customization in deepstream-app