TX2 Install Gesture-Recognition

2019-02-15 更新: NVIDIA Gesture-Recognition 在JetPack-3.3（TX2) 验证tf-openpose 编译Okay

Refer Github link:
NV Gesture-Recognition

Build and install steps:
Requirement :

• Check and install OpenCV
  TX2 Install OpenCV(3.4.0)
• Check and install Tensorflow
  TX2 Install TensorFlow (JetPack3.3)

Start to get nvidia source code

$ git clone --recursive https://github.com/NVIDIA-Jetson/Gesture-Recognition.git

Install program dependencies:

This installs argparse, dill, matplotlib, psutil, requests, scikit-image, scipy, slidingwindow, tqdm, setuptools, Cython, pyTorch, and sklearn

$ sudo pip install -r requirements.txt

Clone the tf-pose-estimation repository:

$ git clone --recursive https://github.com/ildoonet/tf-pose-estimation.git

Configure tf-pose-estimation for our purposes:
This replaces tf-openpose's estimator.py with a custom version for Wave!, and then installs tf-openpose.
Using python2 and pip2 (by dennis)

$ cp -f wavenet/estimator.py tf-openpose/tf_pose/estimator.py
$ cd tf-openpose
$ python setup.py install

If you get errors in building tf_pose like 'tensorpack>0.8.5' and 'fatal error: zmq.h: No such file or directory', please insall pyzmq17.0.0 manually like this:

$sudo pip2 instal pyzmq==17.0.0
......
Collecting pyzmq==17.0.0
tensorpack 0.9.1 requires functools32, which is not installed.
tensorpack 0.9.1 requires msgpack>=0.5.2, which is not installed.
tensorpack 0.9.1 requires msgpack-numpy>=0.4.4.2, which is not installed.
......
 $sudo pip2 instal functools32
 $sudo pip2 instal msgpack
 $sudo pip2 instal msgpack-numpy

Continue...