NVIDIA AGX Xavier 使用记录
硬件信息
NVIDIA AGX Xavier 安装torch和对应torchvision
Jetpack 4.5.1
CUDA 10.2
python=3.6.9
这两个版本比较重要,原因是需要装torch要大于1.7版本的,而大于1.7版本的torch需要装Jetpack,
相关版本,可查看:https://forums.developer.nvidia.com/t/pytorch-for-jetson-version-1-9-0-now-available/72048
对应版本的.whl下载地址:https://elinux.org/Jetson_Zoo#PyTorch_.28Caffe2.29
查看方法可以通过jtop然后下面6INFO来看
jtop安装方式为
# jtop 命令查看cpu和GPU情况
sudo -H pip install jetson-stats
本次安装torch==1.8.0
安装torchvision
git clone -b v0.7.0 https://github.com/pytorch/vision
cd vision
sudo python3 setup.py install
下载onnxruntime
https://elinux.org/Jetson_Zoo#PyTorch_.28Caffe2.29
tensorrt
tensorrt 7.1.3.0
yoloxs 前向推理时间对比
稀疏率:0.5
onnx:
tensorrt:
tensorrt fp6:
tensorrt int8:
稀疏率:0.8
tensorrt fp16:
Reference
https://www.cnblogs.com/stacso/p/15130329.html
https://elinux.org/Jetson_Zoo#PyTorch_.28Caffe2.29
https://blog.csdn.net/Cameron_Rin/article/details/128442082
https://blog.csdn.net/weixin_43653152/article/details/125938135
https://cloud.tencent.com/developer/article/1803089