HUMANSHAPE - 3D Human Shape Modeling, Fitting and Evaluation

This short documentation describes steps necessary to compile and run the 3D human body shape building, manipulation, fitting and evaluation presented in the paper:

Leonid Pishchulin, Stefanie Wuhrer, Thomas Helten, Christian Theobalt and Bernt Schiele
Building Statistical Shape Spaces for 3D Human Modeling In ArXiv, March 2015

http://humanshape.mpi-inf.mpg.de/

Compiling

This code was developed under Linux (Debian wheezy, 64 bit) and was tested only in this environment.

1. Set MATLAB_HOME in following make files:

   external/lbfgsb-for-matlab/Makefile
   shapemodel/Makefile
   evaluation/statQuality/align.mk
   evaluation/statQuality/evaluation.mk
   

2. Switch to the top level directory of the source code, issue make from the command line

Getting the models

1. Download the models

    wget http://datasets.d2.mpi-inf.mpg.de/humanshape/caesar.zip
    wget http://datasets.d2.mpi-inf.mpg.de/humanshape/caesar-norm-wsx.zip
    wget http://datasets.d2.mpi-inf.mpg.de/humanshape/caesar-norm-nh.zip

2. Unzip the models

    unzip caesar.zip && rm -f caesar.zip
    unzip caesar-norm-wsx.zip && rm -f caesar-norm-wsx.zip
    unzip caesar-norm-nh.zip && rm -f caesar-norm-nh.zip

Getting the fitted meshes

1. Download the fitted meshes

    wget http://datasets.d2.mpi-inf.mpg.de/humanshape/caesar-fitted-meshes.zip
    wget http://datasets.d2.mpi-inf.mpg.de/humanshape/caesar-norm-wsx-fitted-meshes.zip
    wget http://datasets.d2.mpi-inf.mpg.de/humanshape/caesar-norm-nh-fitted-meshes.zip

2. Unzip the fitted meshes

    unzip caesar-fitted-meshes.zip && rm -f caesar-fitted-meshes.zip
    unzip caesar-norm-wsx-fitted-meshes.zip && rm -f caesar-norm-wsx-fitted-meshes.zip
    unzip caesar-norm-nh-fitted-meshes.zip && rm -f caesar-norm-nh-fitted-meshes.zip

Running

1. Start matlab
2. Edit file fitting/expParams.m
   1. point p.rootDir to the full path to the source code directory
   2. point p.modelInDir to the model directory, e.g. caesar/
3. Run demo

TODO

Add evaluation code for

1. Per-vertex mean fitting accuracy
2. Total fitting accuracy
3. Compactness, generalization and specificity

If you have any questions, send an email to [email protected] with a topic "humanshape".