ObjectFolder Real

ObjectFolder Benchmark是由斯坦福大学团队创建的多模态物体学习基准数据集，旨在推动多模态感知在计算机视觉、机器人等领域的发展。该数据集包含1,000个神经物体和100个真实世界的家庭物品，涵盖视觉、听觉和触觉三种模态。其中，神经物体通过模拟生成，而真实物体的数据则通过专业的采集流程获取，包括3D网格、视频、冲击声和触觉读数。数据集的创建过程涉及多模态数据的同步采集与校准，确保了数据的高质量与一致性。ObjectFolder Benchmark可用于物体识别、重建和操控等任务，通过多模态数据的融合，揭示不同模态在不同任务中的作用，为多模态物体学习提供了标准化的测试平台。

ObjectFolder Benchmark is a multimodal object learning benchmark dataset created by the Stanford University team, aiming to advance the development of multimodal perception in fields such as computer vision and robotics. This dataset contains 1,000 neural objects and 100 real-world household items, covering three modalities: vision, audition and touch. Among them, neural objects are generated via simulation, while data of real objects are obtained through a professional collection process, including 3D meshes, videos, impact sounds and tactile readings. The dataset creation process involves synchronous acquisition and calibration of multimodal data, ensuring high data quality and consistency. The ObjectFolder Benchmark can be used for tasks such as object recognition, reconstruction and manipulation. Through the fusion of multimodal data, it reveals the roles of different modalities in different tasks, providing a standardized testbed for multimodal object learning.

搜集汇总

背景与挑战

背景概述

ObjectFolder Real数据集包含100个真实家居物体的多感官测量数据，涵盖视觉、听觉和触觉信息，旨在支持多感官物体中心学习的研究。