通过镜像环境安装mmdetection
遇到的问题:
1、没有cuda
解决:在环境中指定CUDA
2、GCC版本太低
解决:升级GCC到5.3版本
dockerfile 如下:
FROM repo.jd.local/public/das:pytorch1.1.0-py3-gpu
# 维护者信息, 或其它标签信息
LABEL maintainer=rengang@jd.com
# CUDA 环境
ENV FORCE_CUDA 1
RUN yum install -y tar
# RUN yum -y install centos-release-scl
# RUN yum -y install devtoolset-7-gcc devtoolset-7-gcc-c++ devtoolset-7-binutils
# RUN echo "source /opt/rh/devtoolset-7/enable" >>/etc/profile
# 升级GCC版本
RUN wget https://copr.fedoraproject.org/coprs/hhorak/devtoolset-4-rebuild-bootstrap/repo/epel-7/hhorak-devtoolset-4-rebuild-bootstrap-epel-7.repo -O /etc/yum.repos.d/devtoolset-4.repo
RUN yum install devtoolset-4-gcc devtoolset-4-gcc-c++ devtoolset-4-binutils -y
RUN echo "source /opt/rh/devtoolset-4/enable" >>/etc/profile
# RUN yum -y install centos-release-scl
# RUN yum -y install devtoolset-6-gcc devtoolset-6-gcc-c++ devtoolset-6-binutils
# RUN echo "source /opt/rh/devtoolset-6/enable" >>/etc/profile
# RUN wget https://ftp.gnu.org/gnu/gcc/gcc-5.2.0/gcc-5.2.0.tar.bz2 &&\
# tar -jxvf gcc-5.2.0.tar.bz2 && \
# cd gcc-5.2.0 && \
# ./contrib/download_prerequisites && \
# mkdir gcc5.2 \
# cd gcc5.2 \
# make -j8 \
# make install \
# 2.a 通过pip安装软件包
RUN source activate base \
&& export PIP_CONFIG_FILE= \
&& pip install --no-cache-dir -i https://pypi.tuna.tsinghua.edu.cn/simple \
imagecorruptions\
matplotlib\
# mmcv>=0.2.15\
mmcv\
numpy\
pycocotools\
six\
terminaltables\
&& rm -rf ~/.cache/pip/*
# RUN git clone https://github.com/open-mmlab/mmcv.git \
# && cd mmcv \
# && pip install -e . \
# 从github上面拉取源码进行安装,以keras官方提供的方法为例
RUN git clone https://github.com/open-mmlab/mmdetection.git \
&& cd mmdetection \
&& source /etc/profile \
&& python setup.py develop \
来源:oschina
链接:https://my.oschina.net/u/4328928/blog/3315537