long inference time to the first images in the folder

[SJavad]

Env

• GPU, Jetson TX1
• OS, Ubuntu 18.04
• Cuda 10.2
• TensorRT 8.2

About this repo

• yolov5
• model: Yolov5

Your problem

@wang-xinyu

• what is your command? e.g. sudo ./yolov5 -d model.engine
• Do you know what your output is? -> inference time 328ms , \n inference time 25ms
• what output do you expect? I expect all photos to take the same amount of time to process
  when I try to infer multiple pictures in the folder, the first and second picture takes a long time to infer but other pictures take the same inference time (35ms)
  what's the problem?
  all pictures have the same dimension. (640,640,3)
  Would anyone be able to help me?

[SJavad]

Has anyone encountered this problem yet?!
Please can I ask you to guide me!? @wang-xinyu

[wang-xinyu]

@SJavad This is as expected, the trt engine needs to warmup.

[SJavad]

@SJavad This is as expected, the trt engine needs to warmup.

So how can I warm up the model?

[wang-xinyu]

By running inference on several images.

If the process started again, it needs to warmup again.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.