Awesome

ICT-FaceKit

ICT's Vision and Graphics Lab's morphable face model and toolkit

ICT Face Model Light

The light version of the ICT Face Model consists of a base topology along with definitions of facial landmarks, rigid, and morphable vertices, and a set of linear shape vectors in the form of principal components of light stage scan data registered to a common topology.

ICT-FaceKit is released under the MIT license.

Face Model Topology

UV Layout

Face Area Details

Eyeball Details

Additional eye geometry including lacrimal fluid, blend meshes, and occlusion meshes adopts the style of Unreal Engine's Digital Human Project. The existing geometries are plug and play with Unreal Engine's shaders.

Teeth Details

Facial Landmarks

All the following indices are 0-indexed.

Multi-PIE 68 point facial landmarks indices:

1225, 1888, 1052, 367, 1719, 1722, 2199, 1447, 966, 3661, 4390, 3927, 3924, 2608, 3272, 4088, 3443, 268, 493, 1914, 2044, 1401, 3615, 4240, 4114, 2734, 2509, 978, 4527, 4942, 4857, 1140, 2075, 1147, 4269, 3360, 1507, 1542, 1537, 1528, 1518, 1511, 3742, 3751, 3756, 3721, 3725, 3732, 5708, 5695, 2081, 0, 4275, 6200, 6213, 6346, 6461, 5518, 5957, 5841, 5702, 5711, 5533, 6216, 6207, 6470, 5517, 5966

Right jawline (substitutes 0-7 in Multi PIE):

1278, 1272, 12, 1834, 243, 781, 2199, 1447

Left jawline (substitutes 9-16 in Multi PIE):

3661, 4390, 3022, 2484, 4036, 2253, 3490, 3496

Right extended jawline (superset of right jawline):

1280, 1278, 1275, 1272, 1248, 12, 820, 1834, 1902, 243, 844, 781, 1673, 2199, 801, 1447, 800

Left extended jawline (superset of left jawline):

3041, 3661, 3042, 4390, 3880, 3022, 3085, 2484, 4102, 4036, 3061, 2253, 3466, 3490, 3493, 3496, 3498

Identity shape vectors

ICT Face Model Light includes a set of 100 PCA modes of linear morph targets. The linear morphing affects the full geometry of the face model and is based on light stage facial scan data.

"What is a Linear 3D Morphable Face Model?" (Youtube link):

Expression Shapes

Current expression shapes adopt the naming convention of the Apple ARKit, but with "Left" and "Right" specified with "_L" and "_R". Additionally, we separeate the shapes (browInnerUp_L and browInnerUp_R), and (cheekPuff_L and cheekPuff_R).

Our expression shapes closest relation to FACS units are as follows:

Full ICT Face Model

Additional features are supported in ICT's Full face model. Some features of the full model are still under development. The full model will be released under a different USC specific license. For questions, or to sign up for a note on when it will be ready, contact kallebladin@gmail.com, or haase@ict.usc.edu

Script Package

This software package comes with some example scripts to get started. First, install external python packages:

pip install numpy
pip install openmesh

cd in to the /Scripts directory as some example scripts require this root path.

Run example scripts:

python sample_random.py
python read_identity.py

Outputs will be written in /sample_data_out

Publications

Learning Formation of Physically Based Face Attributes

CVPR 2020 : IEEE/CVF Conference on Computer Vision and Pattern Recognition

Abstract

Based on a combined data set of 4000 high resolution facial scans, we introduce a non-linear morphable face model, capable of producing multifarious face geometry of pore-level resolution, coupled with material attributes for use in physically-based rendering. We aim to maximize the variety of the participants' face identities, while increasing the robustness of correspondence between unique components, including middle-frequency geometry, albedo maps, specular intensity maps and high-frequency displacement details. Our deep learning based generative model learns to correlate albedo and geometry, which ensures the anatomical correctness of the generated assets. We demonstrate potential use of our generative model for novel identity generation, model fitting, interpolation, animation, high fidelity data visualization, and low-to-high resolution data domain transferring. We hope the release of this generative model will encourage further cooperation between all graphics, vision, and data focused professionals, while demonstrating the cumulative value of every individuals' complete biometric profile.

Citing

BibTex:

@misc{li2020learning,
title={Learning Formation of Physically-Based Face Attributes},
author={Ruilong Li and Karl Bladin and Yajie Zhao and Chinmay Chinara and Owen Ingraham and Pengda Xiang and Xinglei Ren and Pratusha Prasad and Bipin Kishore and Jun Xing and Hao Li},
year={2020},
eprint={2004.03458},
archivePrefix={arXiv},
primaryClass={cs.CV}
}