Human Pose Estimation

Env

win10 +RTX 3080, 使用CUDA 11.8 + TensorRT 8.5 GA + Python 3.10

CUDA >= 11.4

https://developer.nvidia.com/cuda-downloads

>nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:41:10_Pacific_Daylight_Time_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

TensorRT >= 8.4

https://developer.nvidia.com/nvidia-tensorrt-8x-download

添加环境变量

...\TensorRT-8.5.1.7\lib

CUDNN

https://developer.nvidia.com/rdp/cudnn-archive

添加dll所在的bin路径到环境变量

转换yolov8模型为engine

按TensorRT官方教程, trtexec使用g++或visual studio编译下。只要前面的环境变量配置没问题即可。

运行

A:\WorkPlace\YOLOv8-TensorRT>python infer-pose.py --engine ./checkpoints/yolov8s-pose.engine --imgs data --show --out-dir outputs --device cuda:0