-
创建python3环境
conda create -n pointpillars python=3.7 anaconda
-
安装Pytorch1.6
# CUDA 9.2 conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=9.2 -c pytorch # CUDA 10.1 conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.1 -c pytorch # CUDA 10.2 conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.2 -c pytorch
-
安装TensorRT 7.1.3.4
1)下载nvidia-tensorrt-7x-download下载tar格式的
2)解压tar xzvf TensorRT-${version}.${os}.${arch}-gnu.${cuda}.${cudnn}.tar.gz
3)添加环境变量到.bashrc:export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:,其中 是TensorRT文件夹中lib 的路径。之后source /.bashrc
4)安装相关文件:cd TensorRT-${version}/python pip3 install tensorrt-*-cp3x-none-linux_x86_64.whl cd TensorRT-${version}/uff pip3 install uff-0.6.9-py2.py3-none-any.whl cd TensorRT-${version}/graphsurgeon pip3 install graphsurgeon-0.4.5-py2.py3-none-any.whl5)安装torch2trt
git clone https://github.com/NVIDIA-AI-IOT/torch2trt cd torch2trt sudo python setup.py install --plugins //报错试试这个python setup.py install
6)安装其他包:
# Cython needs to be installed before pycocotools pip install cython pip install opencv-python pillow matplotlib pip install git+https://github.com/haotian-liu/cocoapi.git#"egg=pycocotools&subdirectory=PythonAPI" pip install GitPython termcolor tensorboard
7)克隆项目:
git clone https://github.com/haotian-liu/yolact_edge.git cd yolact_edge
8)校准数据集及模型下载:
链接:https://pan.baidu.com/s/1MsfajOuRg4rpvzwo2tVFuw 提取码:aaaa
参考文章:https://blog.csdn.net/qq_44166630/article/details/120761371将coco文件放到data文件夹下,权重文件新建weights文件夹放入其中
9)环境搭建完毕
python3 eval.py --trained_model=weights/yolact_edge_resnet50_54_800000.pth --score_threshold=0.3 --top_k=100 --video=0
报错:RuntimeError: CUDNN_STATUS_INTERNAL_ERROR
解决在eval.py中加入:
torch.backends.cudnn.enabled = False
测试效果:



