RIORefer
收藏资源简介:
RIORefer数据集是由ATR研究机构创建的大型3D视觉定位数据集,旨在克服现有3D视觉定位模型在特定3D数据集上的局限性。该数据集包含超过63,000个多样化的3D对象描述,这些描述分布在1,380个室内RGB-D扫描中,均由人工标注。RIORefer数据集的创建是为了支持Cross3DVG任务,该任务涉及在不同的3D扫描系统上进行3D视觉定位,以评估模型在不同传感器、3D重建方法和语言标注者下的性能。此数据集的应用领域包括增强现实/虚拟现实和个人机器人,旨在通过语言理解实现对真实世界的详细阐释。
The RIORefer dataset is a large-scale 3D visual grounding dataset created by the ATR research institute, aiming to address the limitations of existing 3D visual grounding models on specific 3D datasets. This dataset contains over 63,000 diverse 3D object descriptions, which are distributed across 1,380 indoor RGB-D scans and all are manually annotated. The RIORefer dataset is developed to support the Cross3DVG task, which involves 3D visual grounding across different 3D scanning systems to evaluate model performance across varying sensors, 3D reconstruction methods, and language annotators. The application scenarios of this dataset include augmented reality/virtual reality (AR/VR) and personal robotics, aiming to enable detailed interpretation of the real world through language understanding.

- 1Cross3DVG: Cross-Dataset 3D Visual Grounding on Different RGB-D ScansATR · 2024年



