EFM3D

EFM3D是由Meta Reality Labs Research创建的一个用于评估3D自我中心基础模型进展的基准数据集。该数据集包含约300万个3D定向边界框（OBBs），覆盖43个对象类别，并提供可见性元数据，支持基于图像的检测器训练和评估。EFM3D特别适用于3D对象检测和表面回归任务，旨在推动自我中心机器感知领域的研究。数据集的创建涉及从模拟和真实的自我中心Aria数据中提取和标注信息，确保了数据的真实性和多样性。EFM3D的应用领域主要集中在增强现实和虚拟现实中的空间理解和交互，解决复杂环境下的3D感知问题。

EFM3D is a benchmark dataset developed by Meta Reality Labs Research for evaluating the progress of 3D egocentric foundation models. This dataset contains approximately 3 million 3D oriented bounding boxes (OBBs), covers 43 object categories, and provides visibility metadata to support training and evaluation of image-based detectors. EFM3D is particularly suitable for 3D object detection and surface regression tasks, and aims to advance research in the field of egocentric machine perception. The creation of this dataset involves extracting and annotating information from both simulated and real egocentric Aria data, ensuring the authenticity and diversity of the dataset. The application scenarios of EFM3D are mainly focused on spatial understanding and interaction in augmented reality (AR) and virtual reality (VR), addressing 3D perception challenges in complex environments.

搜集汇总

背景与挑战

背景概述

EFM3D是一个包含约300万个3D定向边界框的基准数据集，覆盖43个对象类别，提供可见性元数据，适用于3D对象检测和表面回归任务。该数据集由Meta Reality Labs Research创建，数据来源于模拟和真实的自我中心Aria数据，主要应用于增强现实和虚拟现实中的空间理解和交互。