DIODE (Dense Indoor and Outdoor Depth)

Diode Dense Indoor/Outdoor Depth (DIODE) 是第一个用于单目深度估计的标准数据集，包括使用相同硬件设置获取的不同室内和室外场景。训练集由 8574 个室内样本和 16884 个室外样本组成，每个样本来自 20 次扫描。验证集包含 325 个室内样本和 446 个室外样本，每组样本来自 10 次不同的扫描。室内训练和验证拆分的地面实况密度分别约为 99.54% 和 99%。户外集的密度自然较低，训练集为 67.19%，验证集为 78.33%。数据集的室内和室外范围分别为 50m 和 300m。Diode Dense Indoor/Outdoor Depth (DIODE) is the first standard dataset for monocular depth estimation, encompassing diverse indoor and outdoor scenes captured using identical hardware configurations. The training set consists of 8,574 indoor samples and 16,884 outdoor samples, with each sample derived from 20 scans. The validation set contains 325 indoor samples and 446 outdoor samples, with each group of samples originating from 10 distinct scans. The ground truth densities of the indoor training and validation splits are approximately 99.54% and 99% respectively. The densities for the outdoor subsets are naturally lower, at 67.19% for the training split and 78.33% for the validation split. The maximum measurement ranges for the indoor and outdoor portions of the dataset are 50m and 300m, respectively.

搜集汇总

背景与挑战

背景概述

DIODE是首个用于单目深度估计的标准数据集，包含室内和室外场景的深度信息，室内训练集样本8574个，室外训练集样本16884个。数据集由北京航空航天大学、芝加哥大学等机构联合发布，发布于2019年。