摘要

我们提供了一个新的可用于单目深度估计测试的稀疏深度标注数据集，包括62帧分辨率为1920*1080的图像及其稀疏深度标注。该数据集对图像中3个平面层次进行深度标注，如图中红框所示，包括花坛侧面、帐篷侧面及建筑侧面。我们通过博世激光测距仪GLM80（测量范围为80m，测量误差为±1.5mm）测量三个平面到相机的距离，花坛侧面到相机距离为45.047m，帐篷侧面到相机距离为32.176m，建筑侧面到相机距离为18.511m。由于激光测距仪的测量误差远小于单目深度估计模型误差，因此我们将其测量结果作为真实值进行稀疏深度标注。

Abstract

We provide a new monocular depth estimation evaluation dataset with sparse depth annotaions. The captured subjects of the dataset are the side of the flower bed, the side of the green tent and the side of the building. We obtain the depth value from the camera to the planes through the laser rangefinder GLM80. The measurement error of GLM80 is ±1.5mm in the measurement range of 80m, which is much lower than that of the estimation from models, thus the measured depth can be viewed as the ground truth. The distance from the flower bed to the camera is 45.047m. The distance from the green tent to the camera is 32.176m. The distance from the building to the camera is 18.511m.

内容

• 62张分辨率为1920*1080的俯拍图像

• 每张图像至少包含三个平面层次中的一个平面

• 对三个平面层次进行稀疏深度标注的mat文件

• 拍摄设备：Canon EOS 80D, 18-135mmIS USM镜头

有关更多详细信息，请参阅数据中包含的README文件。

Contents

• 62 images with 1920*1080 resolution

• Each image contains one or more of the three planes including the side of the flower bed, the side of the green tent and the side of the building.

• The MAT file contains the sparse depth annotation for the three planes.

• Equipment: Canon EOS 80D, 18-135mmIS USM.

数据下载

BHDE-v2:

 bhde_depth_v2.zip

使用该数据库请引用以下文章：

Wang, R.; Zou, J.; Wen, J.Z. SFA-MDEN: Semantic-Feature-Aided Monocular Depth Estimation Network Using Dual Branches. Sensors 2021, 21, 5476. https://doi.org/10.3390/s21165476.