HM3D-ABO

HM3D-ABO是由德克萨斯大学奥斯汀分校创建的一个照片级真实感的以物体为中心的数据集，用于多视角3D重建。该数据集结合了Habitat-Matterport 3D场景数据和Amazon-Berkeley Object物体数据，提供了多视角RGB观测、水密网格模型、地面实况深度图和物体掩码。数据集包含1966个物体，分布在500个室内场景中，适用于相机姿态估计和新视角合成等任务。创建过程涉及随机选择场景和物体，确保物理上合理的放置，并通过渲染和后处理筛选高质量的配置。HM3D-ABO旨在解决现有数据集在真实感和标注丰富性方面的不足，推动AR/VR领域的3D重建技术发展。

HM3D-ABO is a photorealistic object-centric dataset created by The University of Texas at Austin for multi-view 3D reconstruction. This dataset combines Habitat-Matterport 3D scene data and Amazon-Berkeley Object datasets, providing multi-view RGB observations, watertight mesh models, ground-truth depth maps, and object masks. It contains 1966 objects distributed across 500 indoor scenes, and is applicable to tasks such as camera pose estimation and novel view synthesis. Its creation process involves randomly selecting scenes and objects, ensuring physically plausible placements, and screening high-quality configurations through rendering and post-processing. HM3D-ABO aims to address the shortcomings of existing datasets in terms of photorealism and annotation richness, and promote the advancement of 3D reconstruction technologies in the AR/VR field.