SPARE3D

SPARE3D是由纽约大学坦登工程学院开发的一个专注于三维空间推理的数据集。该数据集包含10,369个三维CAD对象，主要用于评估智能系统的空间推理能力。SPARE3D通过三种类型的任务（视图一致性推理、相机姿态推理和形状生成推理）来测试系统的空间智能，这些任务难度递增，旨在激发新的数据驱动研究方向。数据集的创建过程涉及自动生成大量具有标准答案的挑战性问题，用于训练和评估。SPARE3D的应用领域主要集中在提升智能机器人在三维世界中的操作效率，尤其是在没有昂贵三维传感器的情况下。

SPARE3D is a 3D spatial reasoning-focused dataset developed by the Tandon School of Engineering at New York University. This dataset contains 10,369 3D CAD objects, and is primarily used to evaluate the spatial reasoning capabilities of intelligent systems. SPARE3D tests the spatial intelligence of systems through three types of tasks: view consistency reasoning, camera pose reasoning, and shape generation reasoning. These tasks increase in difficulty and are designed to inspire new data-driven research directions. The dataset creation process involves automatically generating a large number of challenging questions with standard answers for training and evaluation purposes. The application scenarios of SPARE3D mainly focus on improving the operational efficiency of intelligent robots in the 3D world, especially when expensive 3D sensors are not available.