The repository provides an implementation of In-The-Wild 3D Morphable Models as outlined in our CVPR paper:
and our journal extension:
3D Reconstruction of "In-the-Wild" Faces in Images and Videos --- J. Booth, A. Roussos, E. Ververas, E. Antonakos, S. Ploumpis, Y. Panagagakis S. Zafeiriou. TPAMI 2018
The following topics are covered, each one with it's own dedicated notebook:
- Building an "in-the-wild" texture model
- Creating an expressive 3DMM
- Fitting "in-the-wild" images
- Fitting "in-the-wild" videos
To leverage this codebase users need to independently source the following items to construct an "in-the-wild" 3DMM:
- A collection of "in-the-wild" images coupled with 3D fits
- A parametric facial shape model of identity and expression
And then to use this model, users will need to provide data to fit on:
- "in-the-wild" images or videos with iBUG 68 annotations
Examples are given for working with some common facial models (e.g. LSFM) and it shouldn't be too challenging to adapt these examples for alternative inputs. Just bear in mind that fitting parameters will need to be tuned when working with different models.
- Follow the instructions to install the Menpo Project with conda.
- Whilst in the conda environment containing menpo, run
pip install git+https://github.com/menpo/itwmm
. - Download a copy of the code into your Downloads folder.
- Run
jupyter notebook
and navigate to thenotebooks
directory in the downloaded folder. - Explore the notebooks in order to understand how to use this codebase.