Advances in Human Pose Estimation and 3D Reconstruction

The recent advancements in human pose estimation and 3D reconstruction have seen significant innovations, particularly in the integration of multi-modal data and the use of advanced neural network architectures. The field is moving towards more robust and efficient methods that leverage both spatial and temporal information, often combining traditional techniques with modern deep learning approaches. Notably, there is a growing emphasis on the development of models that can generalize well across different datasets and scenarios, addressing the limitations of previous methods that were often dataset-specific. Additionally, the incorporation of implicit functions and diffusion models is becoming more prevalent, enhancing the accuracy and realism of 3D human reconstructions. These developments are paving the way for more versatile and high-fidelity digital human representations, which are crucial for applications in interactive telepresence, AR/VR, and the metaverse.

Noteworthy Papers:

  • The introduction of a multi-modal diffusion model for SMPL pose parameters significantly enhances the applicability and accuracy of human pose estimation.
  • A novel two-view geometry estimation framework using implicit differentiation shows substantial improvements in camera pose estimation tasks.

Sources

Unlabeled Action Quality Assessment Based on Multi-dimensional Adaptive Constrained Dynamic Time Warping

Evaluating the evaluators: Towards human-aligned metrics for missing markers reconstruction

Sim2real Cattle Joint Estimation in 3D point clouds

Multi-modal Pose Diffuser: A Multimodal Generative Conditional Pose Prior

POSE: Pose estimation Of virtual Sync Exhibit system

ARTS: Semi-Analytical Regressor using Disentangled Skeletal Representations for Human Mesh Recovery from Videos

Disambiguating Monocular Reconstruction of 3D Clothed Human with Spatial-Temporal Transformer

Efficient Neural Implicit Representation for 3D Human Reconstruction

Robust Two-View Geometry Estimation with Implicit Differentiation

VoxelKeypointFusion: Generalizable Multi-View Multi-Person Pose Estimation

Built with on top of