面向 3D 高斯泼溅中指代分割的几何感知跨模态图对齐
Geometry-Aware Cross-Modal Graph Alignment for Referring Segmentation in 3D Gaussian Splatting
针对 3DGS 场景中的语言指代分割，用位置感知提示词扩展与几何图-语义图跨模态对齐来抑制跨模态注意力的误差自增强，可直接迁移到任何需要文本-3D 空间定位的 3DGS/NeRF 管线。
原文依据 cross-modal attention is self-reinforcing, causing spatial errors to propagate through the Gaussian field once misalignment occurs
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论指代 3D 分割旨在给定自然语言查询的条件下，在三维场景中定位并分割目标物体，需要对几何结构与语言线索进行联合推理。尽管近期基于 3D 高斯泼溅（3DGS）的进展提升了渲染质量，现有方法仍难以将文本指代在空间上落地，原因在于两个根本性局限：(1) 语言编码器不提供显式的位置先验，削弱了几何关系建模；(2) 跨模态注意力具有自我强化特性，一旦发生错位，空间误差便会通过高斯场传播。为解决该问题，我们提出 GeoCGA，一个几何感知的跨模态图对齐框架，将语言语义与 3DGS 表示相衔接。GeoCGA 引入位置感知的提示词扩展，构建捕捉文本中关系结构的语义-空间图，并构建编码三维拓扑的高斯基几何图。跨模态对齐模块强制两个图之间的几何一致性，从而在跨视角下实现稳定且具空间落地性的对应关系。GeoCGA 持续优于此前的最优方法，在 Ref-LERF 上取得 20.8%、在 LERF-OVS 上取得 5.7%、在 3D-OVS 上取得 1.0% 的平均交并比相对提升。
英文原文摘要 1291 字符
Referring 3D segmentation seeks to localize and segment target objects in a 3D scene given a natural-language query, requiring joint reasoning over geometric structures and linguistic cues. Although recent progress using 3D Gaussian Splatting (3DGS) has improved rendering quality, existing methods still struggle to spatially ground textual references due to two fundamental limitations: (1) language encoders provide no explicit positional priors, weakening geometric relation modeling; and (2) cross-modal attention is self-reinforcing, causing spatial errors to propagate through the Gaussian field once misalignment occurs. To address this, we propose GeoCGA, a geometry-aware cross-modal graph alignment framework that bridges linguistic semantics with the 3DGS representation. GeoCGA introduces position-aware prompt expansion to build a semantic-spatial graph capturing relational structure in text, and constructs a Gaussian-based geometric graph encoding 3D topology. A cross-modal alignment module enforces geometric consistency between the two graphs, enabling stable and spatially grounded correspondence across views. GeoCGA consistently outperforms prior state-of-the-art methods, yielding relative mIoU improvements of 20.8% on Ref-LERF, 5.7% on LERF-OVS, and 1.0% on 3D-OVS.