The Make3D dataset is a monocular depth estimation dataset that contains 534 pairs of RGB-D images (including RGB images and their corresponding depth images). Among them, 400 pairs of images are used for training and 134 pairs of images are used for testing. RGB images have high resolution, while depth maps have low resolution.

The depth value of each frame is collected by the laser radar. Compared with the depth information collected by the Kinect camera, this rangefinder can obtain more accurate depth information of outdoor images and has a larger range. Compared with ordinary depth sensors, it also has the characteristics of high resolution and strong anti-interference ability. Since the depth level distinction in the Make 3D image is not obvious, a color heat map is used to represent its depth estimation result.