FlexAvatarLearning Complete 3D Head Avatars with Partial Supervision @MatthiasNiessner
FlexAvatarLearning Complete 3D Head Avatars with Partial Supervision  @MatthiasNiessner
Uploaded December 2025 | Updated November 2026, 2 days ago
tobias-kirschstein.github.io/flexavatar

We introduce FlexAvatar, a method for creating high-quality and complete 3D head avatars from a single image. A core challenge lies in the limited availability of multi-view data and the tendency of monocular training to yield incomplete 3D head reconstructions. We identify the root cause of this issue as the entanglement between driving signal and target viewpoint when learning from monocular videos. To address this, we propose a transformer-based 3D portrait animation model with learnable data source tokens, so-called bias sinks, which enables unified training across monocular and multi-view datasets. This design leverages the strengths of both data sources during inference: strong generalization from monocular data and full 3D completeness from multi-view supervision. Furthermore, our training procedure yields a smooth latent avatar space that facilitates identity interpolation and flexible fitting to an arbitrary number of input observations.

In extensive evaluations on single-view, few-shot, and monocular avatar creation tasks, we verify the efficacy of FlexAvatar. Many existing methods struggle with view extrapolation while FlexAvatar generates complete 3D head avatars with realistic facial animations.
FlexAvatarLearning Complete 3D Head Avatars with Partial Supervision3D Scanning & Motion Capture: 2. Surface RepresentationsDeleted video3D Scanning & Motion Capture: 1. 3DConceptsTUM AI Lecture Series - The multimodal future: Why visual representation still matters (Saining Xie)3D Scanning & Motion Capture: 6. Non-Rigid DeformationDeleted video3D-MPA: Multi Proposal Aggregation for 3D Semantic Instance Segmentation (CVPR 2020)FaceVR: Real-Time Facial Reenactment and Eye Gaze Control in Virtual RealityDeleted videoIntroduction to Deep Learning - 10. Convolutional Neural Networks Part 2 (Summer 2020)High-Res Facial Appearance Capture from Polarized Smartphone Images (CVPR2023)
Matthias Niessner |

FlexAvatarLearning Complete 3D Head Avatars with Partial Supervision

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER