P-CORE: Self-Supervised Surface Consistency for Point-Based Neural Editing

Recent advancements in neural rendering facilitate high-fidelity 3D scene reconstruction, yet non-rigid shape editing remains challenging. P-CORE introduces a self-supervised method allowing point-based representations to handle large deformations and surface discontinuities without ground truth multi-view images. Similarly, PointGT offers a point-based 3D representation for simultaneous geometry and appearance editing. By combining a deformation-friendly point representation with learned UV mapping, PointGT enables high-resolution texture editing, overcoming limitations found in 3D Gaussian Splatting representations.