Using DeepStream Vs. Not using it

[naserie97]

Is it Better to use DeepStream or not to use it. Like I’ve built my own pipeline in C++ and Python using Tensorrt. Does moving toward using DeepStream will make the processing faster or not? I know that it is less error prone to use DeepStream, and it make the development process faster, however since I have many custom code in the pipeline, it will make the development process harder, so Does it worth it to move toward DeepStream in terms of the processing speed like does using DeepStream will increas the fps? and thank you

[AastaLLL]

Hi,

Deepstream adopts TensorRT as the backend so the inference performance should be similar.

However, it optimizes the camera/display pipeline and removes all the unnecessary memory copies.
Maybe you can try a simple network like ResNet to see if there is a significant perf gain when switching to Deepstream.

Thanks.

[naserie97]

Thank you AastaLLL, I would appreciate it if you could advise me on some NVIDIA tools I can learn like Tensorrt and DeepStream that would help me accelerate my pipelines.

[AastaLLL]

Hi,

You can try our profiler to locate the bottleneck of your pipeline first.

Nsight Systems

Thanks.