RPSeg3D

该数据集是一个大型的3D数据集，包含了超过60,000条指令，每条指令都与相应的地面真实部件分割注释一一对应，这些数据是专门为基于推理的3D部件分割而精心策划的。数据集不仅包含彩色点云，还有渲染的2D图像，用于训练模型，使其能够根据隐含的文本查询对3D物体的各个部分进行分割。该数据集的规模超过60,000条指令，其任务是针对3D物体的推理式部件分割。

This is a large-scale 3D dataset meticulously curated for reasoning-based 3D part segmentation tasks. It contains over 60,000 instruction-annotation pairs, with each instruction paired to its corresponding ground-truth part segmentation annotation. The dataset provides both colored point clouds and rendered 2D images, which are used to train models to segment individual parts of 3D objects based on implicit textual queries. This dataset, which encompasses more than 60,000 instruction samples, is dedicated to reasoning-driven part segmentation for 3D objects.