Advances in 3D Human Reconstruction and Animation

The field of 3D human reconstruction and animation is rapidly advancing, with a focus on developing more efficient, accurate, and realistic methods. Recent research has explored the use of end-to-end networks, anatomy shaping, and twins negotiating reconstruction to improve the quality of 3D human avatars. Additionally, there has been a significant emphasis on developing methods for real-time full-body pose estimation, dynamic human neural fields, and compositional TV show reconstruction. These advances have the potential to revolutionize applications such as AR/VR, virtual try-ons, and video production. Noteworthy papers in this area include Unify3D, which introduces a novel paradigm for holistic 3D human reconstruction, and SSD-Poser, which proposes a lightweight and efficient model for robust full-body motion estimation. HumMorph is also noteworthy for its ability to render dynamic human bodies with explicit pose control from few views.

Sources

Unify3D: An Augmented Holistic End-to-end Monocular 3D Human Reconstruction via Anatomy Shaping and Twins Negotiating

SSD-Poser: Avatar Pose Estimation with State Space Duality from Sparse Observations

HumMorph: Generalized Dynamic Human Neural Fields from Few Views

ShowMak3r: Compositional TV Show Reconstruction

AnimateAnywhere: Rouse the Background in Human Image Animation

CompleteMe: Reference-based Human Image Completion

Creating Your Editable 3D Photorealistic Avatar with Tetrahedron-constrained Gaussian Splatting

EfficientHuman: Efficient Training and Reconstruction of Moving Human using Articulated 2D Gaussian

Advance Fake Video Detection via Vision Transformers

MagicPortrait: Temporally Consistent Face Reenactment with 3D Geometric Guidance

Real-Time Animatable 2DGS-Avatars with Detail Enhancement from Monocular Videos

Built with on top of