SERF: Fine-Grained Interactive 3D Segmentation and Editing with Radiance Fields

• DOI: 10.48550/arxiv.2312.15856
• Publication Date: 2023-01-01
• Authors: Kaichen Zhou; Lanqing Hong; Enze Xie; Yongxin Yang; Zhenguo Li; Wei Zhang
• Keywords: Representation; Interactive Visualization

Although significant progress has been made in the field of 2D-based interactive editing, fine-grained 3D-based editing remains relatively unexplored. This limitation can be attributed to two main challenges: lack an efficient 3D representation robust different modifications and absence effective segmentation method. In this paper, we introduce a novel algorithm with radiance fields, which refer as SERF. Our method entails creating neural mesh by integrating multi-view algorithms pre-trained 2D models. Building upon representation, surface rendering technique that preserves local information is deformation. Moreover, forms basis for achieving accurate without requiring supervision. Harnessing facilitates range operations, encompassing tasks such geometry texture painting. Extensive experiments visualization examples on both real synthetic data demonstrate superiority our quality ability.