Skip to content

Latest commit

 

History

History
110 lines (87 loc) · 4.14 KB

INSTALL.md

File metadata and controls

110 lines (87 loc) · 4.14 KB

Setup Locally with Pre-built Wheel

We provide pre-built wheel for python >= 3.4, Ubuntu >= 14.04 or CentOS >=7. The wheels are staticly linked so no dependency other than CUDA is required.

Downdload wheel for CUDA-9.0(2019.8.20)

Downdload wheel for CUDA-10.0(2019.12.14)

Downdload wheel for CUDA-10.1(2019.12.14)

Install the wheel as

pip3 install https://1dv.aflat.top/mxnet_cu101-1.6.0b20191214-py2.py3-none-manylinux1_x86_64.whl --user

Setup with Docker

We provide pre-built docker images for both cuda9.0 and cuda10.0.

Maxwell, Pascal, Volta and Turing GPUs are supported.

For nvidia-driver >= 410.48, cuda10 image is recommended.

For nvidia-driver >= 384.81, cuda9 image is recommended.

Aliyun beijing mirror is provided for users pulling from China.

nvidia-docker run -it -v $HOST-SIMPLEDET-DIR:$CONTAINER-WORKDIR rogerchen/simpledet:cuda9 zsh
nvidia-docker run -it -v $HOST-SIMPLEDET-DIR:$CONTAINER-WORKDIR rogerchen/simpledet:cuda10 zsh
nvidia-docker run -it -v $HOST-SIMPLEDET-DIR:$CONTAINER-WORKDIR registry.cn-beijing.aliyuncs.com/rogerchen/simpledet:cuda9 zsh
nvidia-docker run -it -v $HOST-SIMPLEDET-DIR:$CONTAINER-WORKDIR registry.cn-beijing.aliyuncs.com/rogerchen/simpledet:cuda10 zsh

Setup with Singularity

We recommend the users to adopt singualrity as the default environment manager to minimize the efforts of configuration. Singularity is a virtual environment manager like virtualenv, but in the system-level.

Install Singularity >= 2.6

# install dependency
sudo apt update
sudo apt install build-essential python libarchive-dev

# install singularity
wget https://github.com/sylabs/singularity/releases/download/2.6.1/singularity-2.6.1.tar.gz
tar xzfv singularity-2.6.1.tar.gz
cd singularity-2.6.1
./configure --prefix=/usr/local
make
sudo make install

Download singularity image for SimpleDet

wget https://1dv.aflat.top/simpledet.img

Invoke simpledet shell

Here we need to map the working directory into singularity shell, note that symlink to files outside the working directory will not work since singularity has its own filesystem. Thus we recommend users to map the whole data storage into singularity by replacing $WORKDIR by something like /data or /mnt/.

sudo singularity shell --no-home --nv -s /usr/bin/zsh --bind $WORKDIR /path/to/simpledet.img

Setup from Scratch

System Requirements

  • Ubuntu 16.04
  • Python >= 3.5

Install CUDA, cuDNN and NCCL

Install cocotools

# Install a patched cocotools for python3
pip3 install 'git+https://github.com/RogerChern/cocoapi.git#subdirectory=PythonAPI'

Install MXNet

# Install dependency
sudo apt-get update
sudo apt-get install -y build-essential git
sudo apt-get install -y libopenblas-dev
git clone --recursive https://github.com/apache/incubator-mxnet /tmp/mxnet && \
git clone https://github.com/Tusimple/simpledet /tmp/simpledet && \
git clone https://github.com/RogerChern/cocoapi /tmp/cocoapi && \
cp -r /tmp/simpledet/operator_cxx/* /tmp/mxnet/src/operator && \
mkdir -p /tmp/mxnet/src/coco_api && \
cp -r /tmp/cocoapi/common /tmp/mxnet/src/coco_api && \
cd /tmp/mxnet && \
echo "USE_SIGNAL_HANDLER = 1" >> ./config.mk && \
echo "USE_OPENCV = 0" >> ./config.mk && \
echo "USE_MKLDNN = 0" >> ./config.mk && \
echo "USE_BLAS = openblas" >> ./config.mk && \
echo "USE_CUDA = 1" >> ./config.mk && \
echo "USE_CUDA_PATH = /usr/local/cuda" >> ./config.mk && \
echo "USE_CUDNN = 1" >> ./config.mk && \
echo "USE_NCCL = 1" >> ./config.mk && \
echo "USE_DIST_KVSTORE = 1" >> ./config.mk && \
echo "CUDA_ARCH = -gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70" >> ./config.mk && \
rm /tmp/mxnet/src/operator/nn/group_norm* && \
make -j$((`nproc`-1)) && \
cd python && \
python3 setup.py install && \
rm -rf /tmp/mxnet /tmp/simpledet /tmp/cocoapi