yolox_ros_cpp inference speed up.

[fateshelled]

OpenMP was used in the post-processing, but it was slowing down the process.

Changed to not use OpenMP and some fixed.

[Ar-Ray-code]

I'm also planning to check, how much speed improvement have you seen?

[fateshelled]

In my desktop environment ( OpenVINO ( CPU, intel core i5 11400F ) + yolox_nano ), inference total speed ( preprocess + inference + postprocess ) increased by about 1.5~2 times.

TensorRT ( Jetson Nano 4GB ) + yolox_nano did not change speed much.

[Ar-Ray-code]

For my environment, the FPS stability of TensorRT inference was better after the improvement; OpenVINO could not tell the difference.

Before: 33-125fps
After: 83-100fps

(CPU: Core i9 12900KF GPU: RTX 2080Ti)

[fateshelled]

I did some more tests. All tested yolox_nano model.

• Core i7 8550U

  • OpenVINO CPU: Before 17-22FPS, After 20FPS
  • OpenVINO GPU: Before 20-40FPS, After 50-60FPS

• Core i5 11400F + RTX3060

  • OpenVINO CPU: Before 30-120FPS, After 60-160FPS
  • TensorRT: Before 50-140 FPS, After 70-250FPS

Inference speed and its stability also seem to have improved.

[Ar-Ray-code]

Thank you for your PR 👍 .