Tracking/NvDcf of deepstream-5.0 is not working as done by Tracking/NvDcf of deepstream-4.X

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform (Jetson / GPU) : Jetson TX2 - 4GB Ram
• DeepStream Version : DS-5.0
• JetPack Version (valid for Jetson only) : JetPack 4.4
• TensorRT Version**
• NVIDIA GPU Driver Version (valid for GPU only)**

We are moving our application from Deepstream-4.X to Deepstream-5.0,
We are setting up a pipeline similar to below one

    interpipeSrc-1  -> queue -> NvVideoConvert ->  \														       / AppSinkBin-1 
    interpipeSrc-2  -> queue -> NvVideoConvert ->  /	-> NvStreamMux -> NvInfer -> NvTracker -> NvStreamDemux -> \ AppSinkBin-2
    . . . . . . .    . . . .   . . . . . . . .     /                                                              \  . . . .
    interpipeSrc-N  -> queue -> NvVideoConvert -> /														         \ AppSinkBin-N

this pipeline is correctly working with Deepstream-4.X but when using with Deepstream-5.0 we get following assertion

GPUassert_CUFFT: CUFFT_INVALID_DEVICE /dvs/git/dirty/git-master_linux/deepstream/sdk/src/utils/nvdcf/src/modules/NvDCF/NvDCF.cpp 1012

it seems there is something to be adjusted for new tracking … of for new DCF algorithm …
Could it be our tracking/Dcf configuration files require some modfications ?

Can you check what is required at line 1012 of your nvdcf/src/modules/NvDCF/NvDCF.cpp ?

here more traces got during application running

	setting gie batch-size to 2
	setting gie interval to 2
	OK: create_gie_bin_CCM success
	OK: create_tracking_bin_CCM success** INFO: <Create_Core_Components:705>
	OK: Created element 'demuxer'

   ... setting CoreInferencePipe to PLAYING  ....
	gstnvtracker: Loading low-level lib at /mnt/disk/CRS/CCM_AS_CRS/inferNetwork__Tiny_YoloV3/libnvds_nvdcf.so
	gstnvtracker: Batch processing is ON
	gstnvtracker: Past frame output is OFF
	GPUassert_CUFFT: CUFFT_INVALID_DEVICE /dvs/git/dirty/git-master_linux/deepstream/sdk/src/utils/nvdcf/src/modules/NvDCF/NvDCF.cpp 1012
	terminate called after throwing an instance of 'boost::wrapexcept<boost::log::v2_mt_posix::system_error>'
	  what():  Failed to set TLS value: Invalid argument
	terminate called recursively
3

Not sure if you specify a GPU ID - For Jetson it would necessarily be 0. This error may happen if an invalid ID is used.
Also can you provide the tracker properties.

5

ok, … we checked all parameters and all other stuff … and it seems that problem was due to a contemporary presence of low level linked libraries with different version, … it should be due to a not perfect low level mechanism to build our runtime test application … now we avoided this mixing of libraries and it seems that new tracking is working.

In any case, thanks for support
Maurizio.

6

Hello Maurizio, could you elaborate on which libraries mixing threw the error, it looks like I’m facing a similar predicament?