R4DGS：4D高斯泼溅中的指代分割
原标题：R4DGS: Referring Segmentation in 4D Gaussian Splatting
AI 摘要
该论文提出 Referring 4D Gaussian Splatting（R4DGS）任务，要求在动态 4D 场景中根据自然语言指代表达输出随时间变化、集合级的分割掩码。作者构建了 R4D-Bench-QA 基准，覆盖 4D 变化查询、多目标或推理查询、零目标或干扰查询三类挑战。同时提出免训练框架 ReferGaussian，用 MLLM 将复杂查询分解为可跟踪对象短语与结构化约束，在 EntityBank 上完成约束感知的实体级定位。实验显示其在时间目标选择上有所提升，并在无需场景特定语义训练的情况下取得有竞争力的时空定位效果。
正文节选
R4DGS: Referring Segmentation in 4D Gaussian Splatting Abstract. Dynamic representations based on 4D Gaussian Splatting (4DGS) have shown strong capabilities in reconstruction and rendering, yet current methods remain limited in supporting complex natural-language referring in dynamic scenes. Existing 4D Gaussian methods are primarily designed for rendering-oriented dynamic modeling, while current semantic extensions often rely on scene-specific semantic optimization or are evaluated on relative