:fire: FLAME Universe :fire:
FLAME is a lightweight and expressive generic head model learned from over 33,000 of accurately aligned 3D scans. FLAME combines a linear identity shape space (trained from head scans of 3800 subjects) with an articulated neck, jaw, and eyeballs, pose-dependent corrective blendshapes, and additional global expression blendshapes. For details please see the scientific publication.
Code
List of public repositories that use FLAME (alphabetical order).
- BFM_to_FLAME: Conversion from Basel Face Model (BFM) to FLAME.
- DECA: Reconstruction of 3D faces with animatable facial expression detail from a single image.
- EMOCA: Reconstruction of emotional 3D faces from a single image.
- FaceFormer: Speech-driven facial animation of meshes in FLAME mesh topology.
- FLAME_PyTorch: FLAME PyTorch layer.
- flame-fitting: Fitting of FLAME to scans.
- FLAME-Blender-Add-on: FLAME Blender Add-on.
- GIF: Generating face images with FLAME parameter control.
- learning2listen: Modeling interactional communication in dyadic conversations.
- MICA: Reconstruction of metrically accurated 3D faces from a single image.
- neural-head-avatars: Building a neural head avatar from video sequences.
- photometric_optimization: Fitting of FLAME to images using differentiable rendering.
- RingNet: Reconstruction of 3D faces from a single image.
- SAFA: Animation of face images.
- SPECTRE: Speech-aware 3D face reconstruction from images.
- TF_FLAME: Fit FLAME to 2D/3D landmarks, FLAME meshes, or sample textured meshes.
- video-head-tracker: Track 3D heads in video sequences.
- VOCA: Speech-driven facial animation of meshes in FLAME mesh topology.
Datasets
List of datasets with meshes in FLAME topology.
- VOCASET: 12 subjects, 40 speech sequences each with synchronized audio
- CoMA dataset: 12 subjects, 12 extreme dynamic expressions each.
- D3DFACS: 10 subjects, 519 dynamic expressions in total
- LYHM: 1216 subjects, one neutral expression mesh each.
- Stirling: 133 subjects, one neutral expression mesh each.
- Florence 2D/3D: 53 subjects, one neutral expression mesh each.
- FaceWarehouse: 150 subjects, one neutral expression mesh each.
- FRGC: 531 subjects, one neutral expression mesh each.
- BP4D+: 127 subjects, one neutral expression mesh each.
Publications
List of FLAME-based scientific publications.
2022
- Next3D: Generative Neural Texture Rasterization for 3D-Aware Head Avatars
- Visual Speech-Aware Perceptual 3D Facial Expression Reconstruction from Videos
- Realistic One-shot Mesh-based Head Avatars (ECCV 2022)
- Towards Metrical Reconstruction of Human Faces (ECCV 2022)
- Towards Racially Unbiased Skin Tone Estimation via Scene Disambiguation (ECCV 2022)
- Neural Emotion Director: Speech-preserving semantic control of facial expressions in “in-the-wild” videos (CVPR 2022)
- RigNeRF: Fully Controllable Neural 3D Portraits (CVPR 2022)
- I M Avatar: Implicit Morphable Head Avatars from Videos (CVPR 2022)
- Neural head avatars from monocular RGB videos (CVPR 2022)
- Learning to Listen: Modeling Non-Deterministic Dyadic Facial Motion (CVPR 2022)
- Simulated Adversarial Testing of Face Recognition Models (CVPR 2022)
- EMOCA: Emotion Driven Monocular Face Capture and Animation (CVPR 2022)
- Generating Diverse 3D Reconstructions from a Single Occluded Face Image (CVPR 2022)
- Accurate 3D Hand Pose Estimation for Whole-Body 3D Human Mesh Estimation (CVPR-W 2022)
- MOST-GAN: 3D Morphable StyleGAN for Disentangled Face Image Manipulation (AAAI 2022)
2021
- MorphGAN: One-Shot Face Synthesis GAN for Detecting Recognition Bias (BMVC 2021)
- SIDER : Single-Image Neural Optimization for Facial Geometric Detail Recovery (3DV 2021)
- SAFA: Structure Aware Face Animation (3DV 2021)
- Learning an Animatable Detailed 3D Face Model from In-The-Wild Images (SIGGRAPH 2021)
- Monocular Expressive Body Regression through Body-Driven Attention (ECCV 2020)