MrNeRF (@janusch_patas)
2026-01-13 | โค๏ธ 78 | ๐ 13
ObjSplat: Geometry-Aware Gaussian Surfels for Active Object Reconstruction
Contributions:
-
A geometry-aware view evaluation pipeline that accurately quantifies reconstruction quality and completeness for non-closed surfaces. It explicitly models back-face visibility and true multi-view covisibility, providing reliable guidance for both exploration and refinement.
-
A next-best-path (NBP) planning strategy that performs multi-step lookahead on a dynamically constructed spatial graph. It jointly optimizes information gain and movement cost to generate efficient scanning trajectories, significantly reducing total path length and scan time compared to conventional greedy NBV approaches.
-
A unified active object reconstruction framework, ObjSplat, which leverages surface normals for the incremental update of Gaussian surfels and joint geometryโtexture optimization. This tight coupling enables the efficient, autonomous production of physically consistent, high-fidelity digital assets ready for downstream tasks.
-
Extensive experiments on objects with diverse geometric and textural complexities demonstrate that ObjSplat consistently outperforms existing methods in both reconstruction quality and exploration efficiency. Real-world experiments on four cultural heritage artifacts further validate its robustness and practical applicability.
๋ฏธ๋์ด
๐ Related
Auto-generated - needs manual review