Deepstream 5.0 nvinfer, nvtracker

• Hardware Platform Jetson
• DeepStream Version 5.1
• JetPack Version 4.4.1
• TensorRT Version 7
• questions

I have a deepstream pipeline with only detection. the speed of detection network is 20 fps. if input video/camera has higher speed (like 50fps) what happens to frames? are some of them gonna be dropped or pipe put them in a queue and wait for finishing detection job which lead to latency start to increasing by passing the time?

What latency?

i updated the question. tnx in advance

latency is between the time frames come out of v4l driver queue and the time detection has been executed on frame.

The performance of the pipeline is decided by all components in the pipeline.

For nvinfer, it will not drop any frames. The sink will drop frames which comes too late to be handled.

yes you’re right but I didn’t get my answer.
imagine each frame processing in pipeline takes 40 ms. and camera is 30Hz.
after 10 sec pipe has (10*(40-1/30)ms) delay.
after 20 sec pipe has (20*(40-1/30)ms) delay.
after 30 sec pipe has (30*(40-1/30)ms) delay.

so the delay is increasing during the time. although pipeline has queue but its maximum size is constant and can’t handle all these delayed frames. so the only assumption is that frames are dropping. Is that right???

The frames will be dropped by sink if the property is “sync=TRUE”.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.