README.md

Pix2Face Docker

Instructions for running the pix2face pipeline from within the docker.

Get docker-compose

download docker-compose if you don't already have it, and place it in your PATH.

Clone the source code with submodules

If you didn't use the --recursive flag when cloning, you'll need to run:

git submodule update --init --recursive

Build the docker image

Default build is with GPU support. Use the --cpu-only flag to disable it. This flag is supported by all scripts below.

./docker/build_docker_image.bsh [--cpu-only]

Download the data

./docker/download_data_docker.bsh

Build the software within the docker

./docker/build_pix2face_sources.bsh [--cpu-only]

Run bash inside a named pix2face container

./docker/run_pix2face_interactively.bsh [--cpu-only] [INSTANCE_NAME]

Rendering (i.e. anything that uses face3d.mesh_renderer) can in principle work without the GPU, but currently fails to create a valid OpenGL context. Use the gpu version for now if you need to render images.

Examples: Pose and Coefficient Estimation

After a succseful build these two examples should work

./docker/run_pose_estimation_example.bsh [--cpu-only]
./docker/run_coeff_estimation_example.bsh [--cpu-only]

Pose Estimation

./docker/run_pose_estimation.bsh <image_dir> <output_dir> [--cpu-only]

A single csv file (poses.csv) will be written to <output_dir> containing yaw, pitch, and roll for each image in <image_dir>

Coefficient Estimation

./docker/run_coeff_estimation.bsh <image_dir> <output_dir> [--cpu-only]

One coefficients file per image in <image_dir> will be written to <output_dir>

Jupyter/IPython Notebook

You can run the pix2face demo notebook on port 8885 of your machine with the command:

./docker/run_pix2face_notebook.bsh [--cpu-only]