Advances in 3D Scene Understanding and Reconstruction

The field of 3D scene understanding and reconstruction is rapidly advancing, with a focus on developing more efficient and effective methods for scene representation, object segmentation, and reconstruction. Recent research has explored the use of Gaussian Splatting (3DGS) and other techniques to improve the accuracy and realism of 3D scene reconstruction. Notable developments include the use of reinforcement learning for bottom-up part-wise reconstruction, semantic-driven adaptive Gaussian splatting for extended reality, and geometry-aware assisted depth completion for transparent and specular objects. These innovations have significant implications for applications such as robotics, autonomous systems, and augmented reality. Noteworthy papers in this area include Utilizing Reinforcement Learning for Bottom-Up part-wise Reconstruction of 2D Wire-Frame Projections, which demonstrates the potential of iterative RL wire-frame reconstruction in two dimensions, and SAGE: Semantic-Driven Adaptive Gaussian Splatting in Extended Reality, which effectively reduces memory and computational overhead while keeping a desired target visual quality. Additionally, GAA-TSO: Geometry-Aware Assisted Depth Completion for Transparent and Specular Objects proposes a geometry-aware assisted depth completion method that outperforms other state-of-the-art methods.

Sources

Utilizing Reinforcement Learning for Bottom-Up part-wise Reconstruction of 2D Wire-Frame Projections

SAGE: Semantic-Driven Adaptive Gaussian Splatting in Extended Reality

GAA-TSO: Geometry-Aware Assisted Depth Completion for Transparent and Specular Objects

Is there anything left? Measuring semantic residuals of objects removed from 3D Gaussian Splatting

GS-LTS: 3D Gaussian Splatting-Based Adaptive Modeling for Long-Term Service Robots

SceneSplat: Gaussian Splatting-based Scene Understanding with Vision-Language Pretraining

PanopticSplatting: End-to-End Panoptic Gaussian Splatting

PanoGS: Gaussian-based Panoptic Segmentation for 3D Open Vocabulary Scene Understanding

Online 3D Scene Reconstruction Using Neural Object Priors

MATT-GS: Masked Attention-based 3DGS for Robot Perception and Object Detection

COB-GS: Clear Object Boundaries in 3DGS Segmentation Based on Boundary-Adaptive Gaussian Splitting

Semantic Consistent Language Gaussian Splatting for Point-Level Open-vocabulary Querying

Mitigating Trade-off: Stream and Query-guided Aggregation for Efficient and Effective 3D Occupancy Prediction

Segment then Splat: A Unified Approach for 3D Open-Vocabulary Segmentation based on Gaussian Splatting

Built with on top of