DINK安装视频教程: http://fp-ai.com/video_details.html?id=072b030ba126b2f4b2374f342be9ed44
DINK一键启动视频教程: http://fp-ai.com/video_details.html?id=7f39f8317fbdb1988ef4c628eba02591&type=0
Github:https://github.com/FPAI/DINK
建议采用
* Ubuntu16.04 LTS * 8GB以上内存 * 至少30GB硬盘 * NVIDIA GTX GeForce GPU GTX1060TI以上
1 NVIDIA Docker安装
1.1 准备
* 安装CUDA * 系统设置-->软件与更新-->下载选择其他-->在弹框中选择中国-->选择mirrors.aliyun.com-->右下角选择服务器 * System Settings-->Software &Updates-->Download from-->Other..-->China-mirrors.aliyun.com-->Choose Server-->Close-Reload
1.2 DOCKER CE安装
1.2.1 更新
sudo apt-get update sudo apt-get upgrade
1.2.2 脚本安装docker
curl -fsSL get.docker.com -o get-docker.sh sudo sh get-docker.sh --mirror Aliyun
当以下命令显示含有hello world字样时说明DOCKER环境安装
sudo docker run hello-world
1.3 下载nvidia-docker
wget http://pn7d72sxw.bkt.clouddn.com/nvidia-docker_1.0.1-1_amd64.deb
如果上述下载失败,下载以下
wget https://github.com/NVIDIA/nvidia-docker/releases/download/v1.0.1/nvidia-docker_1.0.1-1_amd64.deb
安装
sudo dpkg -i nvidia-docker_1.0.1-1_amd64.deb
1.4 检查nvidia-docker服务是否存在
systemctl list-units --type=service | grep -i nvidia-docker
如果上述操作失败,运行以下
systemctl list-units --type=service | grep -i nvidia-docker-plugin
安装modprobe
sudo apt-get install nvidia-modprobe
2 下载运行DINK镜像
2.1 拉取镜像前先设置镜像加速器
sudo mkdir -p /etc/docker sudo tee /etc/docker/daemon.json <<-'EOF'
{
"registry-mirrors": ["https://9pbu6dtx.mirror.aliyuncs.com"]
}
EOF sudo systemctl daemon-reload sudo systemctl restart docker
2.2 从阿里云上拉去docker镜像
sudo docker pull registry.cn-hangzhou.aliyuncs.com/dink_framework/dink0.22:latest
2.3 验证镜像:
sudo docker images
显示 registry.cn-hangzhou.aliyuncs.com/dink_framework/dink0.22 latest 即拉取成功
2.4 运行镜像并生成容器用于可视化操作:
下载运行run.sh脚本
wget http://po1ez3p80.bkt.clouddn.com/run_dink6.sh sh run_dink6.sh
2.5 在镜像中启动DINK:
进入容器后直接运行各个.sh脚本文件:
./node_deeplabv3seg_cluster.sh #deeplabv3seg_cluster一键启动节点 ./node_squeezeseg_cluster.sh #squeezeseg_cluster一键启动节点 ./node_voxlelnet.sh #voxlelnet一键启动节点 ./dl_deeplabv3seg_train.sh #deeplabv3seg一键训练 ./dl_deeplabv3seg_eval.sh #deeplabv3seg一键评估 ./dl_squeezeseg_train.sh #squeezeseg一键训练 ./dl_squeezeseg_eval.sh #squeezeseg一键评估 ./dl_voxelnet_train.sh #voxelnet一键训练 ./dl_voxelnet_eval.sh #voxelnet一键评估 ./run_clion.sh #一键启动clion ./run_pycharm.sh #一键启动pycharm (./node_euclidean_cluster.sh #euclidean cluster一键启动节点)