2019-12-29 mmdetection 入坑指南

1.环境

python 3.7

2.安装指导，缺少对nccl的安装指导

https://github.com/open-mmlab/mmdetection/blob/master/docs/INSTALL.md

3.第一次安装运行，是这个样子，失败

export CUDA_HOME=/usr/local/cuda

export PATH=/usr/local/cuda-9.0/bin:$PATH

export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64:$LD_LIBRARY_PATH

export LIBRARY_PATH=/usr/local/cuda/lib64:$LIBRARY_PATH

export PYTHON=$PYTHON:/home/v-zhiwwa/HOI/mmlab/mmdetection/build/lib.linux-x86_64-3.7

cd $mmlab/mmdetection/demo && python wzw.py

4.基础知识

查看pytorch的版本号

import torch

torch.__version__

pytorch对应cuda版本

import torch

torch.version.cuda

查看nvcc的版本

nvcc -V

which nvcc

5.第二次安装，先安装的

cd usr/local && sudo ln -s cuda-10.0 cuda

export PATH=/usr/local/cuda/bin:$PATH

export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

python 3.7

pytorch 1.2.0 支持的cuda 10.0.130

创建nccl安装包

git clone https://github.com/NVIDIA/nccl.git

cd nccl

make -j src.build

make src.build CUDA_HOME=/usr/local/cuda

make -j src.build NVCC_GENCODE="-gencode=arch=compute_70,code=sm_70"

sudo apt install build-essential devscripts debhelper fakeroot

make pkg.debian.build

ls build/pkg/deb/

安装nccl

sudo dpkg -i build/pkg/deb/libnccl-dev_2.5.6-2+cuda10.0_amd64.deb

sudo dpkg -i build/pkg/deb/libnccl2_2.5.6-2+cuda10.0_amd64.deb

安装nccl-test

git clone https://github.com/NVIDIA/nccl-tests.git

cd nccl-tests

make CUDA_HOME=/usr/local/cuda

nccl 2.5.6

gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.12)

python setup.py develop

pip install -v -e .

测试

python

from mmdet.apis import init_detector, inference_detector, show_result

报错，删除build，重新来

6.环境进入

第一次进入存在问题，并解决

tmux new -s mmlab

conda activate open-mmlab

export PATH=/usr/local/cuda/bin:$PATH

export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

export CUDA_HOME=/usr/local/cuda

conda install pytorch==1.2.0 torchvision==0.4.0 cudatoolkit=10.0 -c pytorch

第二次进入，正常进入

tmux new -s mmlab

conda activate open-mmlab

export PATH=/usr/local/cuda/bin:$PATH

export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

export CUDA_HOME=/usr/local/cuda

export PYTHONPATH=$mmlab/mmdetection:$PYTHONPATH