How to use Deepstream's efficient pipeline just for encoding and decoding of rtsp feed?

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hey Guys,

I’m trying to set up IP cameras and before running the inference with deepstream, I want to test just the performance of encoding/decoding part in Deepstream. The input source would be multiple IP cameras, rtsp feed. How do I do that?

Thanks

2

Hi,
You may start with running deepstream-app with sample config file. You can try

deepstream-4.0/samples/configs/deepstream-app$ deepstream-app -c source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_nano.txt

And modify to RTSP source. You can refer to
https://devtalk.nvidia.com/default/topic/1068443/deepstream-sdk/-getting-segmentation-fault-when-running-gst-dsexample-on-multiple-source/post/5411980/#5411980
If you don’t need frame rendering, you can use fakesink

Documents are in
https://docs.nvidia.com/metropolis/deepstream/dev-guide/index.html#page/DeepStream%2520Development%2520Guide%2Fdeepstream_app_config.3.2.html%23
https://docs.nvidia.com/metropolis/index.html

3

Hey DaneLLL,

Thank you. I was able to run this. But it still runs the whole inference pipeline. I want to just use the encoding/decoding part of the pipeline? Read raw input from an IP camera and encode/decode and display as an rtsp output. I’m trying to check the performance of the pipeline in deepstream. Any suggestions to achieve this?

4

Hi,
Please disable [osd], [primary-gie], [tracker] and try again.

5

Thank you DaneLLL.

Do you know where I can find the source to perf measurement?
I can see deepstream prints frames per second, wanted to know how its calculated.

6

Hi,
Please check the code in

deepstream_sdk_v4.0.2_jetson\sources\apps\apps-common\src\deepstream_perf.c