Mosh++ runner

This is a docker container to make running Mosh++ easier!

Download dependencies

Download all of the following dependent files and place them in /downloads.

• bpy-2.83
• smpl-fast-derivative
• pretrained SOMA model for the SOMA dataset
• SMPL-X locked head body model
• extra smplx data
• SSM head marker covariances
• SOMA dataset's manually labeled mocaps (Optional)

Run Mosh++

create docker image

$ docker build -f ./docker/Dockerfile -t moshpp-runner .
$ docker run --privileged -it moshpp-runner bash

inside container

$ bash setup.sh
$ python run_moshpp.py

run_moshpp.py

option	description	default
--mocap_base_dir, -m	Path to the directory containing motion capture data. It is assumed that subdirectories for individual subjects exist directly under this directory, with .c3d files located within those subdirectories.	"~/MOSHPP_WORK_BASE/support_filesevaluation_mocaps/original/SOMA_manual_labeled"
--work_base_dir, -w	Path to the directory where results from Mosh++ will be saved.	"~/MOSHPP_WORK_BASE/running_just_mosh"
--support_base_dir	Path to the directory containing support_files required for running Mosh++.	"~/MOSHPP_WORK_BASE/support_files"
--max_num_jobs	Number of .c3d files to process with Mosh++.	1
--subject	Specify a single subject to process if you want to limit execution to one specific individual.	""
--rotation	Specify [x, y, z] rotation angles in degrees to adjust the orientation if needed. The motion capture data assumes the z-axis is up.	"[0,0,0]"
--unit	Specify the unit of measurement if the .c3d files are not in millimeters. "mm", "cm" "m"	"mm"
--cpu_count	Number of CPUs to use for parallel processing.	256