Amazon Berkeley Objects (ABO)

Amazon Berkeley Objects (ABO) 数据集是由加州大学伯克利分校和亚马逊合作创建的大型数据集，旨在弥合真实世界与虚拟3D世界之间的差距。该数据集包含147,702个产品列表，对应398,212个独特的目录图像，以及每个产品最多18个独特的元数据属性（如类别、颜色、材料、重量、尺寸等）。此外，ABO还包括8,222个产品的“360º视图”旋转台式图像和7,953个产品对应的艺术家设计的3D网格。ABO的3D模型具有复杂的形状和高分辨率的物理基础材料，允许进行照片级真实感渲染。该数据集的应用领域包括单视图3D重建、材料估计和跨域多视图对象检索，旨在解决当前3D计算机视觉方法在真实世界对象上的局限性。The Amazon Berkeley Objects (ABO) Dataset is a large-scale collaborative dataset co-developed by the University of California, Berkeley and Amazon, aimed at bridging the gap between the real world and the virtual 3D realm. This dataset contains 147,702 product listings, corresponding to 398,212 unique catalog images, as well as up to 18 unique metadata attributes for each product (such as category, color, material, weight, dimensions, etc.). In addition, ABO also provides 360° rotating desktop images for 8,222 products and artist-designed 3D meshes for 7,953 products. The 3D models in ABO feature complex shapes and high-resolution physically based materials, allowing for photorealistic rendering. Application scenarios of this dataset include single-view 3D reconstruction, material estimation, and cross-domain multi-view object retrieval, and it is designed to address the limitations of current 3D computer vision methods when applied to real-world objects.

搜集汇总

背景与挑战

背景概述

Amazon Berkeley Objects (ABO) 数据集是一个大规模、多模态的开源数据集，包含147,702个亚马逊产品的多语言元数据、高分辨率目录图像、360度图像序列和高质量3D模型，适用于3D对象理解和计算机视觉研究。数据集提供丰富的几何和材质信息，并采用CC BY 4.0许可证，支持在AWS云上直接访问，具有实际应用和学术研究价值。