Skip to content

Latest commit

 

History

History
185 lines (110 loc) · 6.82 KB

INSTALLATION.md

File metadata and controls

185 lines (110 loc) · 6.82 KB
Raw

Installation of the OVMF

We provide two ways of installing OVMF on your computer:

  1. Easy method: Step-by-step installation that uses a binary package created by us. This method works with Windows only.

  2. Advanced method: Advanced installation including preparation of Blender and compilation of OpenFace. This works for Windows and Linux systems.

In any case, be sure to read about the configuration of your webcam.

1. Easy Installation for Windows

Please follow the instructions step-by-step.

1.1 Install software libraries

Please download and install the following packages. The setups might tell you that the packages are already installed. In this case, re-installation is typically not necessary.

1.2 Setup Conda environment

Open the application called Anaconda Prompt from the start menu. It will typically open in your user's home directory, e.g., c:\Users\mgrewe.
You can change to a different directory or just install the OVMF there. Please make sure that the path does not include spaces. All files will be automatically created in a sub-directory called ovmf.

To prepare and activate a conda environment, type the following lines into the Anaconda Prompt:

conda create -n ovmf python=3.9
conda activate ovmf
conda install git git-lfs

We tested with Python version 3.9, but the OVMF should also run on newer versions (please report any issues).

1.3 Clone OVMF repository

Now, clone the git repository by execution of:

git clone https://github.com/mgrewe/ovmf.git

1.4 Update Conda environment

Install required libraries with:

cd ovmf
conda env update -f .\environment.yml -n ovmf

1.5 Download Blender/OpenFace package

Download the ZIP package and extract it into ovmf\contrib. It should now contain two directories, i.e., Blender and OpenFace.

1.6 Start OVMF

Now, everything should be readily prepared such that you can start the OVMF via:

python start.py -p example

The OVMF should start a Blender and a PsychoPy window showing an example scene like this

Example Scene

Learn how to obtain the FexMM avatars here.

2. Advanced Installation

This walk-through installation will show you how to set up the OVMF with an original version of Blender and the source code of OpenFace. This method has been tested on Windows and Linux.

2.1 Install compiler and build tools

We will now set up a working build tool chain for the compilation of OpenFace.

Windows

Download and run Visual Studio 2019 Build Tools. Select Desktop Development with C++ and then at least MSVCv142 - VS 2019 C++ x64/x86 build tools and Windows 10 SDK for installation. For more information check this guide.

Linux

Install the essential build tools. E.g., in Ubuntu simply run:

sudo apt-get install build-essential

2.2 Install Conda

Download and install Miniconda 64-Bit. Alternatively, you can use full Anaconda 64-Bit. For more information check this guide.

2.3 Setup Conda environment and clone OVMF repository

Create a Conda environment for the OVMF and activate it:

conda create -n ovmf
conda activate ovmf

If necessary, install git, git-lfs, and curl:

conda install git git-lfs curl

Recursively clone the repository and its submodules:

git clone --recurse-submodules https://github.com/mgrewe/ovmf.git
cd ovmf

Install the required libraries in the conda environment. We also installed PsychoPy such that we can start the experiments directly with the OVMF. If it is not needed, you can remove the line from the ovmf/environment.yml. Then run:

conda env update -f environment.yml

2.4 Configure and build OpenFace

Download pretrained detector models (In Anaconda Powershell Prompt, you need to write curl.exe):

curl -L https://www.dropbox.com/s/7na5qsjzz8yfoer/cen_patches_0.25_of.dat -o contrib/OpenFace/lib/local/LandmarkDetector/model/patch_experts/cen_patches_0.25_of.dat

curl -L https://www.dropbox.com/s/7na5qsjzz8yfoer/cen_patches_0.35_of.dat -o contrib/OpenFace/lib/local/LandmarkDetector/model/patch_experts/cen_patches_0.35_of.dat

curl -L https://www.dropbox.com/s/7na5qsjzz8yfoer/cen_patches_0.50_of.dat -o contrib/OpenFace/lib/local/LandmarkDetector/model/patch_experts/cen_patches_0.50_of.dat

curl -L https://www.dropbox.com/s/7na5qsjzz8yfoer/cen_patches_1.00_of.dat -o contrib/OpenFace/lib/local/LandmarkDetector/model/patch_experts/cen_patches_1.00_of.dat

Make sure do download the model files before the build such that they can be copied to the correct locations. Then, configure and build Openface:

cd contrib/OpenFace
mkdir build 
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
cmake --build . --config Release --parallel 4 --target FeatureExtractionPython
cd ../../..

2.5 Install Blender

Download the portable version of Blender. Preferably, extract and move the files into the ovmf/contrib/Blender directory.

Blender comes with its own Python distribution. We need to manually install pyzmq for the communication with the OVMF. This can be done by opening a terminal in the Blender directory and executing the following lines provided that the version is changed accordingly:

Windows

.\contrib\Blender/3.0\python\bin\python.exe -m pip install -U pip
.\contrib\Blender/3.0\python\bin\python.exe -m pip install pyzmq

Linux

./contrib/Blender/3.0/python/bin/python3.9 -m pip install -U pip
./contrib/Blender/3.0/python/bin/python3.9 -m pip install pyzmq

2.6 Configure OVMF

OVMF needs to know the location of Blender and OpenFace. Adjust the paths in ovmf/config/contrib.json accordingly. Below the OVMF directory, relative paths can be used, e.g.,

{
    "blender_renderer_executable": "contrib/Blender/blender",
    "openface_binary_dir": "contrib/OpenFace/build/bin/",
    "openface_model_au_root_dir": "contrib/OpenFace/build/bin/"
}

2.7 Start OVMF

Now, everything should be setup. Run the example as described here.