Question about output

AI & Data Science Deep Learning (Training & Inference) TensorRT
1
  • I apologize in advance that the grammar is not accurate because I am Korean.

Description

Hi. I’m using the code YOLOv5 and the converted TenserRT from YOLOv5.
Overall, TensorRT engine is faster than YOLOv5 model,
but the speed of the prediction is the same.
the speed is only a difference in the nms

I wonder why the speed is difference where the output value is processed, not the inference

1

  1. TensorRT engine

    • prediction : 10 ms
    • nms : 7.5 ms

  2. YOLOv5 model

    • prediction : 9.6 ms
    • nms : 24 ms

Thank you.

Environment

TensorRT Version: 8.6.1.6
GPU Type: rtx 3080 laptop
Nvidia Driver Version: 531.61
CUDA Version: 11.3
CUDNN Version: 8.9.6.50_cuda11
Python Version (if applicable): 3.9.13
PyTorch Version (if applicable): 1.12.0+cu113

2

Hi @mjeonghun ,
Can you please share your model with repro script for us to try .

Thanks