Transfer_learning_app error:

hi,
i followed this topic

and got patch to change as said in topic, but shows erro:
** ERROR: main:841: Failed to set pipeline to PAUSED
Quitting
ERROR from src_bin_muxer: Output width not set
Debug info: /dvs/git/dirty/git-master_linux/deepstream/sdk/src/gst-plugins/gst-nvmultistream/gstnvstreammux.c(2779): gst_nvstreammux_change_state (): /GstPipeline:pipeline/GstBin:multi_src_bin/GstNvStreamMux:src_bin_muxer
App run failed

pls advice me how to fix that erro to run?
and another question, transfer_learning_app can enable kafka to run directly in config file?

thank you very much!

hi,
i make again deepstream-app and transfer_learning_app can runs right now, but where is bbox on images which takes?
adding new converter caps before pgie is necceciary or not? how can i move image_probe after OSD?
and sometimes, that app shows erro but i setup confidence 0.5 and 1.0:

Confidence (-0.1) provided by neural network output is invalid. ( 0.0 < confidence < 1.0 is required.)
Please verify the content of the config files.
**PERF: 14.32 (15.20)
Confidence (-0.1) provided by neural network output is invalid. ( 0.0 < confidence < 1.0 is required.)
Please verify the content of the config files.

Hi,

You might be a similar issue as below.
Please check our document for the confidence = -0.1 issue:
https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_FAQ.html?highlight=confidence#why-do-i-see-confidence-value-as-0-1

Thanks.

thank you maybe tracker problem and i will try again.
what about msg kafka by transfer_learning app?

thank you

hi,
i know deepstream-app has supports msgconv plugin and i may enable that at config file directly, but my model is custom yolo and new classes for objects.
should i define msg informations similar with test5 in deepstream-app again? because when i enable that nothing topic was created at my cloud.
pls confirm for me how to use kafka with transfer learning app together?

thank you

Confidence (-0.1) provided by neural network output is invalid. ( 0.0 < confidence < 1.0 is required.)
the problem is caused by tracker and i have got it to run no error shows.

for deepstream_app, i confirmed that it does not support msgconv plugin directly, even i moved test5 codes which includes all msg informations to deepstream_main_app.c kafka doesn’t work still.
i also tried make transfer learning app with test5 together, but insteresting thing is kafka can run but no image to save… pls help to check where is breaking for image probe???
my project target is using transfer learning app to work kafka together, pls give an advice to solve the matter soon,

thank thank a lots, lots!!

hi,
i have got transfer learning to work image, kafka and rtsp well right now, that’s good sample more than test5…
but one question is kafka informations’s timestamp was lately 8 hours, and i used local time by rfc3399 only.
where can i get it correct time?

thank you

Hi,

Thanks for updating the progress.
Good to know the transfer learning sample and Kafka is working now.

For the time issue, have you applied this workaround again?

Thanks

my system clock is CST time already.
problem is RFC3339 probe to get UTS format from system only.
how can fix the probe to get CST time? that doesn’t support localtime which same as test5, so that’s reason test5 have to make another clock file together.
anyway this is a problem for project apply.
pls advice an easy way to fix RFC3339 to CST…

thank you very much!