This research proposes a system for deformable object reconstruction and segmentation from a single point cloud. This method is beneficial for robot-assisted surgical applications, where the ability to identify and interact with specific segments of a deformable object, such as organs, is crucial. The proposed system uses neural occupancy functions to improve object registration and segmentation. It is tested on a variety of objects in both simulated and real-world scenarios, demonstrating its ability to robustly identify object segments.
Publication date: 13 Nov 2023
Project Page: https://arxiv.org/abs/2311.07357v1
Paper: https://arxiv.org/pdf/2311.07357
