chcorbi/helvipad

Helvipad是一个用于全向立体深度估计的真实世界数据集，包含来自不同环境的40K帧视频序列，涵盖了拥挤的室内和室外场景以及不同的光照条件。数据集通过使用两个360°摄像头和LiDAR传感器收集，并通过将3D点云投影到等距柱状图像上来生成精确的深度和视差标签。此外，还提供了通过深度补全增强的训练集，显著增加了标签密度。

Helvipad is a real-world dataset for omnidirectional stereo depth estimation, consisting of 40K frames from video sequences across diverse environments such as crowded indoor and outdoor scenes. The dataset is collected using two 360° cameras and a LiDAR sensor, providing accurate depth and disparity labels. Additionally, an augmented training set with increased label density through depth completion is provided. The dataset structure is divided into training and testing subsets, containing depth maps, disparity maps, RGB images, and LiDAR point cloud data. This dataset is used to evaluate the performance of various stereo matching methods, particularly for omnidirectional image depth estimation.