This package includes the neural network and part of the code from the open source project OpenFace which has been implemented in C++ for use with this robot project.
The code is supposed to run on a NVIDIA Tegra Jetson TK1 board. The login data for this board are:
Username: ubuntu
Password: ubuntu
In order to compile this library some dependencies are required. Most importantly the OpenFace library depends on dlib, OpenCV and Torch.
To install the torch library, start a new terminal and run the following commands.
$ git clone https://github.com/torch/distro.git ~/torch --recursive
$ cd ~/torch; bash install-deps;
$ ./install.sh
OpenCV has to be version 3.1 and needs to be compiled from source with cuda support. Check this link for a complete description of this step:
$ TODO
At some point it would be nice to have dlib as a cmake submodule but for now you have to download it and put it into the include path manually.
$ cd ~/Downloads
$ wget http://dlib.net/files/dlib-18.18.tar.bz2
$ tar jxf dlib-18.18.tar.bz2
$ cp -R dlib-18.18/dlib /usr/local/include/
$ rm dlib-18.18.tar.bz2
$ rm -r dlib-18.18
Next we also need some lua packages for to run the neural network in torch.
== Possible errors == === Not found: ... errors === If you get a lot of not found: <torch_function_name> errors, then you are compiling against the wrong lua library. So in the CMakeLists.txt file you should find the lua library in target_link_libraries and make sure that it links against luajit.
=== File.lua:370: table index is nil === This error usually comes from trying to load a 64-bit binary neural network on a 32-bit machine. The solution can then be to export the neural network to ascii format on a 64-bit machine.