Skip to content

Latest commit

 

History

History
71 lines (46 loc) · 2.81 KB

installation_jetson_tx2_jetpack3.3.md

File metadata and controls

71 lines (46 loc) · 2.81 KB
Raw

OpenPose Doc - Installation on Nvidia Jetson TX2 JetPack 3.3

Note that OpenPose for Nvidia Jetson TX2 was developed and it is maintained by the community. The OpenPose authors will not be able to provide official support for it.

Contents

  1. Requirements and Dependencies
  2. Installation
  3. Usage

Requirements and Dependencies

Jetson TX2 just flashed with JetPack 3.3

Notes:

  • Installation is similar to Jetson TX1 and you can follow this step by step tutorial.
  • If you are installing from a virtual machine host, installation may need to be done in two steps, please refer to this solution.
  • Be sure to complete both OS flashing and CUDA / cuDNN installation parts before installation.

Dependencies:

- OpenCV (3.X versions are compatible).
- Caffe and all its dependencies.
- The demo and tutorials additionally use GFlags.

Installation

Use the following script for installation of both caffe and OpenPose:

bash ./scripts/ubuntu/install_caffe_and_openpose_JetsonTX2_JetPack3.3.sh

Optional: If you want to build the Python libraries, then:

  1. Edit the BUILD_PYTHON flag on CMakeLists.txt:

option(BUILD_PYTHON "Build OpenPose python." ON)

  1. In both places where this appears, set the flag to ON:
-DBUILD_python=ON
-DBUILD_python_layer=ON
  1. There are additional flags that need to be set: PYTHON_EXECUTABLE=/usr/bin/python2.7 and PYTHON_LIBRARY=/usr/lib/aarch64-linux-gnu/libpython2.7.so for Python 2.7. Therefore, inside build, do:

cmake -DBUILD_PYTHON=ON -DPYTHON_EXECUTABLE=/usr/bin/python2.7 -DPYTHON_LIBRARY=/usr/lib/aarch64-linux-gnu/libpython2.7.so ..

  1. Now run make. You should see a file called "pyopenpose.so" if Python was set to 2.7, in /home/nvidia/openpose/build/python/openpose. Otherwise, it will be pyopenpose.cpython-35m-aarch64-linux-gnu.

  2. Finally, run sudo make install inside build to copy the files to /usr/local/python and set PYTHONPATH accordingly on .bashrc:

export PYTHONPATH="${PYTHONPATH}:/usr/local/python

Usage

It is for now recommended to use an external camera with the demo. To get to decent FPS you need to lower the net resolution:

./build/examples/openpose/openpose.bin -camera_resolution 640x480 -net_resolution 128x96

To activate hand or face resolution please complete this command with the following options (warning, both simultaneously will cause out of memory error):

--hand -hand_net_resolution 256x256
--face -face_net_resolution 256x256