Core Concepts
Entity-NeRF effectively removes moving objects and reconstructs static urban backgrounds in dynamic scenes.
Abstract
1. Introduction
Neural Radiance Fields (NeRF) face challenges in modeling urban scene dynamics.
Entity-NeRF combines knowledge-based and statistical strategies for accurate reconstruction.
2. Related Works
NeRF models struggle with unbounded scenes, adaptations like NeRF++ address this.
3. Preliminaries
Challenges in urban scenes include diverse moving objects and complex backgrounds.
4. Method
Entity-wise Average of Residual Ranks (EARR) identifies moving objects using entity-wise statistics.
5. Results
Evaluation on MovieMap Dataset shows Entity-NeRF outperforms RobustNeRF in foreground PSNR.
6. Conclusion
Entity-NeRF excels in removing moving objects but has limitations with shadows.
Stats
RobustNeRFは、移動オブジェクトを取り除くための純粋な統計的アプローチです。
Entity-wise Average of Residual Ranks(EARR)は、エンティティごとの統計を使用して移動オブジェクトを識別します。