Advances in Human Motion Analysis and Synthesis

The field of human motion analysis and synthesis is rapidly evolving, with a focus on developing more accurate and efficient methods for estimating and generating human motion. Recent research has explored the use of multi-stage avatar generators, prototype-guided fashion video generation, and geometry-level 3D human-scene contact estimation to improve the accuracy and realism of human motion synthesis. Additionally, there has been a growing interest in developing methods for human motion prediction, 3D human reconstruction, and contactless fingerprint recognition. Noteworthy papers in this area include MAGE, which proposes a multi-stage avatar generator for inferring full-body poses from sparse observations, and ProFashion, which introduces a prototype-guided fashion video generation framework for improving view consistency and temporal coherency. GRACE is also a notable paper, which presents a new paradigm for 3D human contact estimation by incorporating a point cloud encoder-decoder architecture and hierarchical feature extraction and fusion module. MTVCrafter is another significant contribution, which proposes a 4D motion tokenization framework for open-world human image animation, achieving state-of-the-art results with an FID-VID of 6.98. Overall, these advancements have the potential to revolutionize various applications, including virtual reality, robotics, and biometrics.

Sources

MAGE:A Multi-stage Avatar Generator with Sparse Observations

ProFashion: Prototype-guided Fashion Video Generation with Multiple Reference Images

GRACE: Estimating Geometry-level 3D Human-Scene Contact from 2D Images

When Dance Video Archives Challenge Computer Vision

Human Motion Prediction via Test-domain-aware Adaptation with Easily-available Human Motions Estimated from Videos

Link to the Past: Temporal Propagation for Fast 3D Human Reconstruction from Monocular Video

Monocular Online Reconstruction with Enhanced Detail Preservation

G-MSGINet: A Grouped Multi-Scale Graph-Involution Network for Contactless Fingerprint Recognition

M3G: Multi-Granular Gesture Generator for Audio-Driven Full-Body Human Motion Synthesis

AfforDance: Personalized AR Dance Learning System with Visual Affordance

EmbodiedMAE: A Unified 3D Multi-Modal Representation for Robot Manipulation

MTVCrafter: 4D Motion Tokenization for Open-World Human Image Animation

ADHMR: Aligning Diffusion-based Human Mesh Recovery via Direct Preference Optimization

Built with on top of