ImputeRefer

该数据集名为ImputeRefer，它包含了从原始ScanNetv2数据集中提取的点云场景，并在此基础上增加了人类指向手势和修改后的参照表达式。该数据集的创建旨在通过融入手势信息来增强现有数据集，同时使用IoU@0.25和IoU@0.5评价指标对其进行评估。数据集规模包括707个场景，共35,581个样本（其中9,391个为测试样本）。该数据集针对的任务是3D-Embodied Reference Understanding（3D-Eru）。

This dataset is designated as ImputeRefer. It consists of point cloud scenes extracted from the original ScanNetv2 dataset, with added human pointing gestures and modified referring expressions. The dataset is developed to augment existing datasets by integrating gesture information, and its performance is evaluated using the IoU@0.25 and IoU@0.5 metrics. It includes 707 scenes with a total of 35,581 samples, 9,391 of which are test samples. The task targeted by this dataset is 3D-Embodied Reference Understanding (3D-Eru).