当前位置:网站首页>[mmdetection] solves the installation problem

[mmdetection] solves the installation problem

2022-07-06 17:23:00 TiggerRun

install

introduction

It has been used again in recent competitions MMdetection library , The last time I used it was in the Tianchi race baseline, This time Ubuntu Reinstall under the server MMdetection Or hit a wall everywhere , Summarize the following questions , A lot of problems are github issue Repeat in .

Create a new environment

conda create -n name python==3.6

Recommended 3.6 edition

cuda edition

#  Look at... On the server cuda edition , It is highly recommended to install 10.x edition !
cat /usr/local/cuda/version.txt

install pytorch

pip install torch==1.1.0 torchvision==0.3.0 cudatoolkit==10.0.130

To avoid mistakes , This is the most stable version configuration

correct numpy edition

 pip install numpy==1.16.0

numpy v1.18 Can cause int and float64 Calculation incompatibility , stay mmdetection In test Calculation AP Failure

MMdet Supported by pytroch edition

mmdetV1.0:torch1.1,1.2,1.3,1.4
mmdetV2.0:torch1.5

it is to be noted that pytorch Version of , There's been a recent putorch1.5, If you do not specify a version to install, it will be installed as pytorch1.5, however 1.5 Version is in MMdetectionV2.0 Compatible in , If needed V2.0 Then use the following statement to install

git clone https://github.com/open-mmlab/mmdetection -b v2.0
cd mmdetection

cython,wheel

pip install cython wheel

mmcv

pip install mmcv

perhaps

 git clone https://github.com/open-mmlab/mmcv.git
 cd mmcv
 pip install -e .

If installed mmcv There is a problem , You can try to use

pip install pytest-runner

This is a necessary library

pycocoapi

 pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI

perhaps

 pip install pycocotools

mmdetection v1

git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetection
pip install -e -v .

After installation failure

python setup.py clean

Delete manually build file , After dealing with the problem, continue with the following

pip install -v -e .

perhaps

python setup.py develop
原网站

版权声明
本文为[TiggerRun]所创,转载请带上原文链接,感谢
https://yzsam.com/2022/187/202207060928240047.html