EG-3DVG：面向 3D 视觉定位的表达式与几何感知定位解码器
EG-3DVG: Expression and Geometry Aware Grounding Decoder for 3D Visual Grounding
面向 3D 视觉定位的解码器设计：位置引导的表达式交叉注意力把文本线索可靠注入视觉词元，几何感知掩码注意力只聚合几何一致的邻域特征，可直接迁移到任何需要文本-点云对齐的 grounding 或指代分割流程。
原文依据 position-guided expression cross-attention (PECA) for reliable text-vision alignment and geometry-aware masked attention (GMA)
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论尽管 3D 视觉定位近期取得了进展，现有方法仍受三个核心挑战困扰：1）跨模态错位导致文本线索无法可靠地传递到视觉表征；2）对细粒度表达式线索理解不足而引发的类内混淆；3）空间相关视觉特征聚合不准确造成的几何推理错误。我们提出 EG-3DVG，一个通过表达式与几何感知定位解码器解决上述问题的统一框架。该解码器集成两个互补的注意力模块——位置引导的表达式交叉注意力（PECA）用于可靠的文本-视觉对齐，几何感知掩码注意力（GMA）用于对几何一致的视觉线索进行选择性聚合。为进一步区分语义相似的实例，我们引入表达式感知对比学习（ECL），强化目标物体词元与表达式相关词语之间的对齐。在 ScanRefer 与 SR3D/NR3D 上的大量实验表明，EG-3DVG 在 3D 边界框定位与掩码预测两方面均达到最优性能，验证了几何感知与表达式感知设计的有效性。
英文原文摘要 1253 字符
Despite recent progress in 3D visual grounding, existing methods still struggle with three core challenges: 1) cross-modal misalignment that prevents textual cues from being reliably delivered to visual representations, 2) intra-class confusion arising from insufficient understanding of fine-grained expression cues, and 3) geometric reasoning errors caused by inaccurate aggregation of spatially relevant visual features. We propose EG-3DVG, a unified framework that addresses these issues through an expression and geometry aware grounding decoder. The decoder integrates two complementary attention modules--position-guided expression cross-attention (PECA) for reliable text-vision alignment and geometry-aware masked attention (GMA) for selective aggregation of geometry-consistent visual cues. To further distinguish semantically similar instances, we introduce expression-aware contrastive learning (ECL), which strengthens the alignment between the target object token and expression-relevant words. Extensive experiments on ScanRefer and SR3D/NR3D demonstrate that EG-3DVG achieves state-of-the-art performance in both 3D bounding box localization and mask prediction, validating the effectiveness of our geometry- and expression-aware design.