• Hardware Platform (Jetson / GPU):Jetson
• DeepStream Version: 5.0
• JetPack Version (valid for Jetson only):4.4
• TensorRT Version: 7.1
I used tlt_model+nvdcf_tracker+custom_model in the Gstreamer pipeline of deepstream.
The latency of these two pipeline are:
1- tlt_model+nvdcf_tracker = 10 ms
2- tlt_model+nvdcf_tracker+custom_model = 20 ms
In the pipeline 1, the NvDCF tracker has good accuracy, but in the pipeline 2 because the latency is increased and this cause to occurred dropping accuracy for tracker and the numbers of ID tracker higher than reality.
The one solution can be to separate the tlt_model+nvdcf_tracker with custom_model in the gstreamer pipeline.
Is it possible to do this with queue element between two parts? How to solve this problem?