Pls give an advice for deepstream:

when i run deepstreamSDK test5 in docker, and show logs like this:

it seems less some lib file, pls give me some advice for this problem, or i have to modify test5.c?

thank you very much!

Did you use original test5 app? can you share commands you used?

i used original test5 to test that, but i changed primegie to custom yolo model, this casued the problem.
does i need to modify test5.c to match my classes, right?

my command is deepstream-test5-app -c xxx.txt, deepstream runs at docker.

thanks

i got the problem logs by:

sudo docker logs NVIDIADeepStreamSDK

and i have another urgent questions to take your time, too:
if i use official standard model as pgie, and i want to use my custom yolo model as sgie, but this model is object detector not a classfier,
can this work well?

thank you very much!

Yes, you need to do some changes according your model.
You can refer to GitHub - NVIDIA-AI-IOT/deepstream_tlt_apps at release/tlt3.0
which support YOLO v3&v4 model.

Yes, we support back to back detectors
you can refer to this sample,

i worked back to back, so it is both detectors, i know now and have to move test5 to back to back…

pls also give an advice after find face, how can i recognition them?
should i work SPIE by classfier, too? had better show a instance to me.

thank you very much!

Do not quite follow, what do you mean “how can i recognition them”, your sgie is a detector, not a classifier?

i mean if i user back_to_back as main procedure, it is only detect object of face and LP, but i want to know whose face,
how can i do next step?

thank you for your time!

yes my sgie is a detector of yolo and not a classifier.
it can work as a sgie on back-to-back?

thanks

and i have another question to ask:
i checked back-to-back instance again, if i use kafka or azure, it supports msgbroker already? or i have to add element in back-to-back instance myself?

thank you very much

and another questions about back-to-back instance, its elements like this:
if (!source || !h264parser || !decoder || !primary_detector || !secondary_detector
|| !nvvidconv || !nvosd || !sink)

if i work two secondary_detector, can i add like this:
if (!source || !h264parser || !decoder || !primary_detector || !secondary_detector_1 || !secondary_detector_2
|| !nvvidconv || !nvosd || !sink)

pls give some advice urgently!
thank you very much

As i know, to know whose face, does that should belong to classification?

->seems not correct, you may implement by comparing face feature vectors to recognize whose face.

Yes.

it does not include nvmsgbroker. if you need, you need to add it into the app.

thank you very much!
and this question:

and another questions about back-to-back instance, its elements like this:
if (!source || !h264parser || !decoder || !primary_detector || !secondary_detector
|| !nvvidconv || !nvosd || !sink)

if i work two secondary_detector, can i add like this:
if (!source || !h264parser || !decoder || !primary_detector || !secondary_detector_1 || !secondary_detector_2
|| !nvvidconv || !nvosd || !sink)

pls give some advice urgently!
thank you very much

theoratically it should be ok to connect 3 detectors, but we did not tried, if you met problem, please file new topic.

ok i will add 3 secondary_sgie to try in my project.
is it possible to give an advice to add msgbroker in pipline of back-to-back because it is a little trouble for test5.app?
i have to work kafka and azure to cloud.

thank you very much!

You can refer to test4 app.