Docker image for FrankMocap

The page of the image on Dockerhub: https://hub.docker.com/r/nickveld/frankmocap-env

Preparation

Packages

In case the usage of this image you can skip conda setup, installing packages using apt-get, conda and pip listed in the installation guide from the repository of frankmocap.

Optional: Building

There is no need to rebuild image but in case of rebuilding you need to place requirements.txt from the docs directory of the repository of frankmocap next to the Dockerfile

Also, you can download supplementary files (the details below) into the image using the following command:

ARG INSTALLATION_PATH="TYPE THE PATH HERE"

WORKDIR ${INSTALLATION_PATH}
COPY scripts scripts
# `find | xargs sed` line is aimed to reduce the number of `wget` progress updates
# because each update causes new line in the `docker build` output
RUN find scripts -name '*.sh' -type f | xargs sed -i 's/wget/wget --progress=dot:giga /g' \
  && sh scripts/install_frankmocap.sh \
  && rm -r scripts

Optional: Building freegult

If you want to build freegult you can you the following lines as a starter for Dockerfile:

RUN apt-get update \
  && apt-get install -y cmake unzip libx11-dev libxi-dev \
  && rm -rf /var/lib/apt/lists/*
WORKDIR /tmp
RUN wget https://sourceforge.net/code-snapshots/svn/f/fr/freeglut/code/freeglut-code-r1865-tags-FG_3_2_1.zip \
  && unzip freeglut-code-r*-tags-FG_*.zip \
  && cd freeglut-code-r*-tags-FG_* \
  && cmake . \
  && make \
  && cd .. \
  && rm -r freeglut-code-r*-tags-FG_*

For other versions:

1. See https://sourceforge.net/p/freeglut/code/HEAD/tree/tags/
2. Select the version and enter into its folder
3. Click on "Download Snapshot"
4. Cancel the downloading and copy the direct link
5. Replace the URL argument of wget with the obtained link

Although, I am not successed with the building because

/tmp/freeglut-code-r1865-tags-FG_3_2_1/include/GL/freeglut_std.h:144:13: fatal error: GL/glu.h: No such file or directory
    include <GL/glu.h>
            ^~~~~~~~~~

I assume that somewhere in the files for CMake libGL library linking is missing.

Optional: pytorch3d

If you need the prebuilt image with pytorch3d replace nickveld/frankmocap-env (the image name) with nickveld/frankmocap-env:pytorch3d. In case building add the build argument PIP_AUX_PKGS=pytorch3d. (Example: docker build --build-arg PIP_AUX_PKGS=pytorch3d)

Supplementary files and packages from git

The installation scripts from the repository of frankmocap is not good for the usage with the docker environments. Thus I have slightly modificated them, see this pull request or this branch. The docker image and this instruction assume that you have the same content of the scripts directory.

After you cloned the repository of frankmocap and applied the changes to the files inside the scripts directory, change your working directory to the root of the repository (the directory with scripts, LICENSE, ...). You can run sh scripts/install_frankmocap.sh either in the host system or in the docker container (see the "Usage" section).

SMPL/SMPLX models

Follow the minisection "Setting SMPL/SMPL-X Models" in the original installation guide.

Filesystem tree validation

After running sh scripts/install_frankmocap.sh and downloading the SMPL/SMPLX models compare your filesystem tree under the root of the repository with the tree in the section "Folder hierarchy" in the original installation guide. In Unix-like systems (including the system inside the docker container) the tree command can assist you.

Usage

Basic run command is simple (assuming that your working directory is the root of the repository (the directory with scripts, LICENSE, ...):

docker run -it --rm -v $(pwd):/opt/app -w /opt/app nickveld/frankmocap-env

Note that the path after -v $(pwd): and the path after -w are the same, and it is important!

If you have not run sh scripts/install_frankmocap.sh but intend to do it in the docker, see the subsection "Troubleshooting" below in order to get details regarding it.

Running without a display

If you want to run a demo from frankmocap you need most likely to use xvfb-run that allows to run the demo without a display. But exec xvfb-run does not work properly somewhy (exec "$@" is a traditional way to execute the provided docker command). For example, we need to run the following command: python -m demo.demofrankmocap YOUR_ARGS. One way is the folowing one:

user@host_system:/path/to/frankmocap$ docker run -it --rm -v $(pwd):/opt/app -w /opt/app nickveld/frankmocap-env
root@docker_container:/opt/app$ xvfb-run -a python -m demo.demofrankmocap YOUR_ARGS

But also a one line way is implemented (using the special environment variable USE_XVFB set with -e USE_XVFB=1):

user@host_system:/path/to/frankmocap$ docker run -it --rm -v $(pwd):/opt/app -w /opt/app -e USE_XVFB=1 nickveld/frankmocap-env python -m demo.demofrankmocap YOUR_ARGS

The image is based on pytorch/pytorch:1.6.0-cuda10.1-cudnn7-runtime which, in turn, is based on nvidia/cuda:10.1-cudnn7-devel-ubuntu18.04 . Thus, refer to their documentation in order to enable GPU in the container and learn other advanced options.

Troubleshooting

• After the run I see "The setup routine is needed, but the setup script is not located..."
  • Check that the path between -v and : points to the repository root ($(pwd) means "the current directory")
  • Check that the path after -v $(pwd): and the path after -w are the same
• After the run I see "python: can't open file 'setup.py': [Errno 2] No such file or directory"
  • If you have not run sh scripts/install_frankmocap.sh you must run it just after you see the error message. (Or run it on the host system as described in the subsection "Supplementary files and packages from git")
  • Check the filesystem tree as described in the subsection "Filesystem tree validation"
• After a demo from frankmocap execution finishes I see Segmentation fault and get exit code 139
  • I do not know how to fix it. I only have figured out that there is a problem with opengl renderer (or its compatibility with xvfb-run). You can just switch to another renderer and the message disappears.