RGB 序列递增式 3D 语义场景图预测（英文版）

英文标题：Incremental 3D Semantic Scene Graph Prediction from RGB Sequences中文摘要：本论文提出一种实时框架，使用 RGB 图像序列增量构建一致的三维语义场景图，包括新颖的增量实体估计流水线和场景图预测网络，使用多视角和几何特征，通过迭代信息传递估计三维语义场景图。经过在 3RScan 数据集上大量实验证明，该方法在这项挑战性任务中表现良好，优于现有的最先进方法。英文摘要：D semantic scene graphs are a powerful holistic representation as theydescribe the individual objects and depict the relation between them. They arecompact high-level graphs that enable many tasks requiring scene reasoning. Inreal-world settings, existing 3D estimation methods pr