Centrala begrepp
NARUTO is a groundbreaking system that combines neural representation with uncertainty learning to achieve high-fidelity surface reconstruction in active environments.
Sammanfattning
The NARUTO system introduces a novel approach to active reconstruction by integrating hybrid neural representations and uncertainty-aware planning. It autonomously explores uncertain observations, enhancing reconstruction completeness and fidelity. The system outperforms existing methods in benchmark datasets like Replica and MP3D.
Statistik
Our system achieves exceptional reconstruction completeness, advancing state-of-the-art performance from 73% to 90%.
Extensive evaluations confirm superior performance in various environments using an indoor scene simulator.
The completion ratio increases as uncertainty decreases, showcasing the effectiveness of the uncertainty-aware planning approach.