Advances in 3D Scene Reconstruction and Dynamic Modeling

The field of 3D scene reconstruction and dynamic modeling is rapidly advancing, with a focus on developing methods that can accurately capture and predict complex, dynamic scenes. Recent research has centered around improving the accuracy and efficiency of 3D reconstruction techniques, such as Gaussian Splatting, and integrating them with dynamic modeling approaches to better handle moving objects and scenes. Notable advancements include the development of novel frameworks that enable real-time reconstruction of dynamic 3D scenes from monocular videos, as well as methods that can accurately estimate 3D motion and trajectory from 2D tracking sequences. Additionally, researchers have made significant progress in applying these techniques to real-world applications, such as surgical scene reconstruction, robotic navigation, and human motion capture. Overall, the field is moving towards more accurate, efficient, and robust methods for 3D scene reconstruction and dynamic modeling. Noteworthy papers include ODE-GS, which introduces a novel method for dynamic scene extrapolation using latent ODEs, and HAIF-GS, which proposes a unified framework for structured and consistent dynamic modeling using sparse anchor-driven deformation. DynaSplat is also notable for its approach to dynamic-static separation and hierarchical motion modeling for scene reconstruction.

Sources

ODE-GS: Latent ODEs for Dynamic Scene Extrapolation with 3D Gaussian Splatting

Layered Motion Fusion: Lifting Motion Segmentation to 3D in Egocentric Videos

Where Is The Ball: 3D Ball Trajectory Estimation From 2D Monocular Tracking

SurGSplat: Progressive Geometry-Constrained Gaussian Splatting for Surgical Scene Reconstruction

Dy3DGS-SLAM: Monocular 3D Gaussian Splatting SLAM for Dynamic Environments

A Real-time 3D Desktop Display

StreamSplat: Towards Online Dynamic 3D Reconstruction from Uncalibrated Video Streams

Design of an innovative robotic surgical instrument for circular stapling

HAIF-GS: Hierarchical and Induced Flow-Guided Gaussian Splatting for Dynamic Scene

DynaSplat: Dynamic-Static Gaussian Splatting with Hierarchical Motion Decomposition for Scene Reconstruction

MetricHMR: Metric Human Mesh Recovery from Monocular Images

Fluoroscopic Shape and Pose Tracking of Catheters with Custom Radiopaque Markers

DGS-LRM: Real-Time Deformable 3D Gaussian Reconstruction From Monocular Videos

Transformer IMU Calibrator: Dynamic On-body IMU Calibration for Inertial Motion Capture

InstaInpaint: Instant 3D-Scene Inpainting with Masked Large Reconstruction Model

Built with on top of