CO3D

由于无法获得真实数据，因此主要仅使用合成数据集来探索学习重建对象类别的3D结构。CO3D通过提供由带有相机姿势和地面真相3D点云注释的对象类别的真实多视图图像组成的大规模数据集，促进了该领域的进步。

CO3D数据集包含来自近19,000个视频的总共150万帧，这些视频捕获了来自50 MS-COCO类别的对象。因此，它在类别和对象的数量上都超过了替代方案。该数据集适用于学习特定类别的3D重建和新视图综合方法，如原始的NeRF。~ 由于真实数据的不可用性，学习重建对象类别的3D结构主要是使用合成数据集进行探索的。CO3D通过提供由用相机姿态和地面真实3D点云注释的对象类别的真实多视图图像组成的大规模数据集，促进了该领域的进步。

CO3D数据集包含来自近19000个视频的150万帧，这些视频捕捉了来自50个MS-COCO类别的对象。因此，它在类别和对象的数量上都超过了其他选择。该数据集适用于学习特定类别的3D重建和新的视图合成方法，如开创性的NeRF。

Given the unavailability of real-world data, research on learning 3D structure reconstruction for object categories has primarily been conducted using synthetic datasets. CO3D has advanced the field by releasing a large-scale dataset composed of real multi-view images of object categories, annotated with camera poses and ground-truth 3D point clouds.

The CO3D dataset comprises a total of 1.5 million frames from nearly 19,000 videos, which capture objects across 50 MS-COCO categories. Consequently, it outperforms alternative datasets in terms of both the count of categories and individual objects. This dataset is well-suited for learning category-specific 3D reconstruction and novel view synthesis methods, such as the original NeRF.

Due to the unavailability of real-world data, learning 3D structure reconstruction for object categories has mainly been explored using synthetic datasets. CO3D has advanced the field by providing a large-scale dataset consisting of real multi-view images of object categories annotated with camera poses and ground-truth 3D point clouds.

The CO3D dataset includes 1.5 million frames from nearly 19,000 videos, which capture objects from 50 MS-COCO categories. As a result, it outperforms other available options in terms of both the number of categories and objects. This dataset is applicable to learning category-specific 3D reconstruction and novel view synthesis methods, such as the pioneering NeRF.

搜集汇总

背景与挑战

背景概述

CO3D是一个大规模的真实多视图图像数据集，包含来自50个MS-COCO类别的近19,000个视频的150万帧，每帧都带有相机姿势和地面真相3D点云注释。该数据集适用于学习特定类别的3D重建和新视图综合方法，如NeRF，推动了该领域的研究进展。