MICA (MetrIC fAce) takes a single in-the-wild image as input and reconstructs 3D face shapes which are metrically more accurate than previous methods.
EMOCA (EMOtion Capture and Animation) takes a single in-the-wild image as input and reconstructs a 3D face with sufficient facial expression detail to convey the emotional state of the input image.
PIXIE (Pixels to Individuals: eXpressive Image-based Estimation) reconstructs an expressive body with detailed face shape and hand articulation from a single image. PIXIE does this by regressing the body, face and hands directly from image pixels using a neural network that includes a novel moderator, which attends to add weights information about the different body parts.
ToFu (Topological consistent Face from multi-view) infers a dense registration mesh for face, directly from the multi-view image input (with known camera calibration), without the need for photogrammetry or mesh registration.
DECA (Detail Expression Capture and Animation) takes a single image as input and outputs a 3D face shape in FLAME mesh topology and animatable details that change with expression. This codebase provides training and inference code.
GIF (Generative Interpretable Faces) is a generative 2D face model with FLAME's parameter control. Given FLAME parameters for shape, pose, expressions, parameters for appearance, lighting, and an additional style vector, GIF outputs photo-realistic face images. This codebase provides training and inference code.
STAR (Sparse Trained Articulated Human Body Regressor) is a generative 3D human body model trained from scans of ~14,000 subjects, with sparse & local pose blend shapes, more realistic deformations and only ~20% of the parameters of SMPL. STAR fixes artifacts and issues with SMPL, reduces parameters, is trained from many more bodies, and is a drop-in replacement for SMPL. This codebase demonstrates how to load STAR in different code frameworks (i.e. Tensorflow, PyTorch, and Chumpy).
ExPose (EXpressive POse and Shape rEgression) is a method that estimates 3D body pose and shape, hand articulation and facial expression of a person from a single RGB image. This codebase contains code to regress an expressive body (i.e. with articulated hands and facial expressions) from a single image.
VOCA (Voice Operated Character Animation) is a simple and generic speech-driven 3D facial animation framework that works across a range of identities. This codebase demonstrates how to synthesize realistic character animations given an arbitrary speech signal and a static character mesh.
This repository contains the code to fit an expressive body (i.e. with articulated hands and facial expressions) to images.
The codebase consists of the inference code, i.e. give an face image using this code one can generate a 3D mesh of a complete head with the face region.
CoMA (Convolutional Mesh Autoencoder) is a versatile model that learns a non-linear representation of a face using spectral convolutions on a mesh surface. This codebase demonstrates how to train a convolutional mesh autoencoder from a registered 3D face dataset.
FLAME (Faces Learned with an Articulated Model and Expressions) is a statistical head model trained from over 33, 000 scans that combines a linear shape space with an articulated jaw, neck, and eyeballs, pose-dependent corrective blendshapes, and additional global expression. This codebase demonstrates how to load and play with FLAME and fit it to scans.
This codebase demonstrates how to fit a 3D Morphable Model to a single image using landmarks and edge features.
Template fitting code for non-rigid mesh registration. This codebase demonstrates how to non-rigidly deform a 3D template mesh to fit a target 3D scan.
This codebase demonstrates how to load global multilinear 3D face models and how to fit it to a face scan.
This codebase demonstrates how to load a multilinear wavelet 3D face models and how to fit it to noisy or partially occluded 3D face scans.
This codebase demonstrates how to load global and local 3D face models and how to fit them to noisy or partially occluded 3D face scans.
A robust framework to robustly learn a multilinear model from 3D face databases with missing data, corrupt data, wrong semantic correspondence, and inaccurate vertex correspondence. Further, a multilinear face model learned from the joint BU-3DFE and Bosphorus databases is provided.
Groupwise multilinear correspondence optimization method, and different optimized multilinear 3D face models.