Running deepstream-text1 on tx2 to load yolov5s engine model becomes very delayed

[autodrive2022]

Please provide complete information as applicable to your setup.
• Jetson tx2
• DeepStream Version 6.0
• The code language is c++
1, deepstream-test1 comes with a traditional model that runs very smoothly on tx2
2, I replaced the engine model loaded by deepstream-text1 to yolov5s, but I found that running deepstream-text1 on tx2 to load the yolov5s model became very delayed.
3, Test: I use deepstream-app which comes with deepstream to run the engine for yolov5 and it is very fast.
4, Question: I need to adjust those properties of element in the pipleline of deepstream-text1 to improve the running rate of engine model running yolov5s?

[fanzh]

please set model-engine-file after the first time. please refer to sample config config_infer_primary.txt.

[autodrive2022]

Yes, I have commented out yolov5s cfg and wts configuration items in config config_infer_primary.txt, leaving only the engine configuration items. As a result, it became fast to start, but the video latency was long and felt like a dropped pin.

[fanzh]

if you developed your own DS C++ code, you could also refer to DeepStream SDK FAQ - #12 by bcao to measure the latency of the pipeline comonents

[autodrive2022]

I used deepstream-test1, just modified the code-general configuration file to config_infer_primary.txt

[fanzh]

1. can deepstream-test1 run ok without any modification?
2. the information is not enough, please measure the latency of the pipeline comonents according to DeepStream SDK FAQ - #12 by bcao, like this: Comp name = primary_gie in_system_timestamp = 1656325320313.372070 out_system_timestamp = 1656325320323.925049 component latency= 10.552979

[autodrive2022]

Thank you for your answer.
Yes，deepstream-test1 without any modification.
The deepstream-app loading yolov5s engine model is ok.
I guess it might have something to do with the fact that deepstream-test1 is loading a custom model like yolov5s and its pipeline itself is not perfect
I will think about it.

[fanzh]

Any further update? Is this still an issue to support? Thanks

[autodrive2022]

Thanks!
Resolved！