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

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