Pi3D, HEB

Pi3D数据集由苏黎世联邦理工学院计算机视觉与几何组创建，包含约1000个3D平面，这些平面是从1DSfM数据集的10000张图像中观察到的。HEB数据集是一个大规模的单应性估计基准，利用Pi3D数据集，包含226,260个单应性变换，用于评估各种鲁棒估计器和基于深度学习的对应关系过滤方法。这两个数据集广泛应用于单目深度训练、表面法线估计和图像匹配算法的评估，以及鲁棒单应性估计的研究。数据集创建过程中，使用了COLMAP进行场景重建，确保了数据集的高质量和适用性。应用领域包括计算机视觉、增强现实和自动驾驶等。

The Pi3D Dataset was developed by the Computer Vision and Geometry Group at ETH Zurich. It contains approximately 1,000 3D planes observed from 10,000 images sourced from the 1DSfM dataset. The HEB Dataset is a large-scale homography estimation benchmark. Built upon the Pi3D Dataset, it encompasses 226,260 homographic transformations, which are utilized to evaluate various robust estimators and deep learning-based correspondence filtering methods. These two datasets have been extensively applied in the evaluation of monocular depth training, surface normal estimation, image matching algorithms, as well as research on robust homography estimation. During the dataset creation process, COLMAP was employed for scene reconstruction, ensuring the high quality and applicability of the datasets. Its application fields include computer vision, augmented reality, autonomous driving, and other related areas.