OpenPose Installation
Published: 2019-05-15

I saw the introduction of OpenPose on the Internet and felt it was a good thing, although the current hardware does not meet its requirements and cannot be verified.However, this time the installation process was translated for later use.

OpenPose installation:

Operating System Support

Ubuntu 14 and 16. Windows 8 and 10. Nvidia Jetson TX2 ,Installation Guide See doc/ . OpenPosehas also been used Windows 7 ,
Mac, CentOS,
and Nvidia Jetson (TK1 and TX1) Embedded System .However ,We do not formally support them at present. .

Hardware Requirements

NVIDIAvideo card, at least available 1.6 GB (Ubuntuupper command nvidia-smidetection available GPUmemory ). at least 2 GBmemory . recommended cuDNNand at least one 8of stone CPU.

These requirements are considered the default configuration (such as --net_resolution
"656x368" and
scale_number 1).
You may need more.
Or less.

Download Update Resource Library

The first step is to download the OpenPose library.
It may be put on GitHub of windows and Ubuntu terminals.

git clone

OpenPose can be easily updated by clicking on windows GitHub
Top right corner of Desktop
The synchronization button,
Or in Ubuntu
Inside runs gitpul origin master.
In OpenPose
After downloading,
Run on your operating system only what is described as Reinstallation.
Part of.




Recommended installation method.
It is relatively simple and provides more customized settings.
Note that this is a beta version.
If it fails,
Please come in to GitHub inside and use Installation at the same time.
- Script Compilation .


And Atlas must already be installed on your machine:

1. [CUDA](
Must be installed.
You should restart your machine after installing CUDA.

2. [cuDNN](
Once you have downloaded it,
Decompress it and merge the content into the CUDA folder, such as ` /usr/local/cuda-8.0/ `.
Note: we found that in cuDNN
5.1 OpenPose Run ~10%
Faster than on cuDNN 6.
Otherwise, check [compilation
without cuDNN](#compiling-without-cudnn).

3. OpenCV uses`apt-get
If you have compiled OpenCV 3 yourself, refer to [Manual
After the Makefile.config file is generated,
Edit them and delete the comment line ` # OPENCV_VERSION := 3'.
Or you can modify all Makefile.config.UbuntuXX'
File and run step2 script.

In addition, OpenCV 3
The' opencv_contrib' module is not included by default.
If you have OpenCV with contrib module
3 and you want to use it in the' Makefile' file inside's' LIBRARIES'
+= opencv _ coreopencv _ higgui opencv _ imgproc' insert opencv_contrib' after line.

5. Atlas uses sudo
Instead of Atlas, you can modify this line of BLAS'
: = atlas use OpenBLAS
Or Intel MKL.


Building CAFFE (Convolutional Architecture for Fast Feature Embedding) & OpenPose
Library, download Ubuntu 14.04 or 16.04 required CAFFE (Convolutional Architecture for Fast Feature Embedding) module (script automatic detection)
And CUDA 8:

bash ./ubuntu/

This script is only available in CUDA 8
And Ubuntu 14 or 16.
Otherwise see also
- Manual Compilation.


Scripted installation is selected.
If you want to use CUDA 7,
Avoid sh
Modify some configuration labels (such as OpenCV

installation Caffe
. compilation Caffeand OpenPose,Run the following command line :

3. ### Install
Caffe ###

4. cd 3rdparty/caffe/

5. # Select
your desired Makefile file(run only one of the next 4 commands)

6. cpMakefile.config.Ubuntu14_cuda7.example
Makefile.config # Ubuntu 14, cuda 7

7. cpMakefile.config.Ubuntu14_cuda8.example
Makefile.config # Ubuntu 14, cuda 8

8. cpMakefile.config.Ubuntu16_cuda7.example
Makefile.config # Ubuntu 16, cuda 7

9. cp Makefile.config.Ubuntu16_cuda8.exampleMakefile.config
# Ubuntu 16, cuda 8

10.# Change
any custom flag from theresulting Makefile.config (e.g. OpenCV 3, Atlas/OpenBLAS/MKL, etc.)

11.# Compile
CAFFE (Convolutional Architecture for Fast Feature Embedding)

12.make all
-j${number_of_cpus}&& make distribute -j${number_of_cpus}


14.### Install
OpenPose ### ../../models/

16.bash ./
# It justdownloads the Caffe trained models ..

18.# Same file
cp command as the oneused for Caffe


20.# Change
any custom flag from theresulting Makefile.config (e.g. OpenCV 3, Atlas/OpenBLAS/MKL, etc.)

21.make all

If you want to use your own distributed CAFFE (Convolutional Architecture for Fast Feature Embedding),
According to the following in Custom
CAFFE (Convolutional Architecture for Fast Feature Embedding)
Part of the steps, then recompile the OpenPose library:

bash ./

These steps need only be performed once.
If you are interested in making changes to the OpenPose library, you can simply compile it:

make clean

make all -j$(NUM_CORES)

is very important:
There are 2 makefiles.config.ubuntu # # .example
Similar documents,
One is under the main document, the other is under the 3rdparty/caffe/ CAFFE (Convolutional Architecture for Fast Feature Embedding)/,
Equivalent to OpenPose
With CAFFE (Convolutional Architecture for Fast Feature Embedding)
The configuration file for.
Any changes must be made to 2 files at the same time.
(e.g. opencv3flag, atlab/openblas/mklflag, etc.).
For example, for CUDA 8 and Ubuntu16:
as well as


If you update some software (our library or a third-party library used), or you simply want to reinstall it:

Empty OpenPoseand CaffeCompile folder :

make clean && cd 3rdparty/caffe && make clean

repetition Installation step .You don't need to download these modules anymore .


You only need to delete the OpenPose folder.
For example, rm -rf openpose/.




Download and decompress portable
OpenPose demo 1.0.1


Prerequisites for Installation : Microsoft VS2015 Enterprise Update 3. VS2017 Communitynot supported , VS 2015 Communityhas not been tested . CUDA 8 :Install to Default Location , C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0 .Otherwise ,corresponding modification Visual Studio project VS2015After installation, install CUDA 8.0,ensure CUDAinstall build pair VSall necessary documents .If CUDAis already installed ,in installation VSReinstall After CUDA! cuDNN 5.1 :Once you have downloaded it ,unzip the copy (merge)Its content to CUDAfolder , C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0 . download OpenPoseDependencies and Modules (body, face and hand models),double click {openpose_path}\windows\download_3rdparty_and_models.bat .is equivalent to ,You can download them manually : module : COCO
: download in models/pose/coco/ . MPI
: download in models/pose/mpi/ . Face
: download in models/face/ . Hands
: download in models/hand/ . dependency : Caffe :
Unzip as
3rdparty/windows/caffe/ . Caffe
: Unzip as 3rdparty/windows/caffe3rdparty/ . OpenCV
: Unzip as 3rdparty/windows/opencv/ . on VS project slnfile, i.e. double click {openpose_path}\windows\OpenPose.sln . For confirmation OpenPoseis running ,Attempt to compile for execution demo: right key OpenPoseDemo --> Set as StartUp Project . to read Debug by Release mode. for F5Compile Run . If you have already connected webcam, OpenPosewill start automatically after compilation . To use the created from the command line exefile ,You must : copy {openpose_folder}\3rdparty\windows\caffe\bin\ofAll DLLsto {openpose_folder}\windows\x64\Release . copy {openpose_folder}\3rdparty\windows\opencv\x64\vc14\bin\ofAll DLLsto : {openpose_folder}\windows\x64\Release . on cmd (Windows button + X, then A). Enter OpenPosedirectory : cd C:\openpose\ . Run tutorial commands . in the default images, video or webcamrun on OpenPose,to verify the correct installation OpenPose: doc/ .