This project produces a .deb installation package for Torch. The motivation
is that there were many cases where I had to go through the whole Torch build
process to run OpenFace in a new environment, and I wanted to speed the setup
of that environment.
To build a workable package, it is best to execute the build inside the container specified by the Dockerfile in the project root directory.
To build the Docker image, change to this project directory and execute
$ docker build -t torch-deb .
To build the .deb file, you can run the default container command and then
copy the output file from the container filesystem to your host filesystem,
or you can run a bash shell and build from within the container.
Execute
$ docker run --mount type=bind,src=/tmp,dst=/mnt torch-deb
to build the .deb and copy it to /tmp. Change src=/tmp to point to a
different destination directory if you so please.
To start a shell, execute
$ docker run --rm -it torch-deb /bin/bash -l
To build the .deb from within that container, execute
$ mvn -f /root/torch-parent/pom.xml install -Pcontainer
You can do this on your host computer or in a Docker container. For a suitable container, execute
$ docker run --rm -it ubuntu:18.04 /bin/bash -l
You can also just use the torch-deb container to avoid having to install all
the dependencies.
OpenFace setup is as follows:
# apt update
# apt install --yes ./torch-private_7.0-1_all.deb python-pip wget git cmake
# git clone https://github.com/cmusatyalab/openface.git
# cd openface
From here on, you might prefer to use an Anaconda or virtualenv environment.
To create an Anaconda environment, execute conda create --name openface python=2.7.
After entering the environment (conda activate openface), continue as follows.
# pip install -r requirements.txt
# pip install dlib==19.16.0 opencv-python==4.0.0.21
# models/get-models.sh
At this point, your OpenFace is functional. To demonstrate, execute this:
# PYTHONPATH=. demos/compare.py images/examples/{lennon-*,clapton-*}