There is a delay between the RTSP video stream output by Deepstream and the source RTSP video stream

When i am using yolov5m.engine for target detection in Deepstream. There is a delay between the output RTSP video stream and the source RTSP video stream, and this delay increases with the running time, when running for 9 hours, the delay reaches 3 minutes.

Environment:
Jetson Xavier NX
Deepstream-5.0

Any idear?Thanks.

We can’t know anything with this description.

I try this instruction:
#deepstream-app -c /deepstream_app_config_yoloV5.txt
Then the output video stream has a delay of about 15 seconds
The longer the running time, the greater the delay.

You use yolov5 maybe very big network for Xavier

So it’s nvinfer’s bottleneck, you should notice the fps and gpu usage.

And maybe u should try other model precision(fp16 int8) or model pruning

Thanks for your reply,and i am using this project [tensorrtx/yolov5 at master · wang-xinyu/tensorrtx · GitHub].The model I use is fp16 and it works well. However, there is a continuously increasing delay when running in deepstream.

You should confirm the fps reached 25 to makes it work without no delay, since it’s a live stream. You can also set interval of nvinfer plugin to more than zero, so the resource usage could decrease a lot to reach the live streaming.

1 Like

I have solved the problem very well using your ideas, thank you very much.

hi how did you do, I have the same problem. It takes very long time to detect