Depth map super-resolution method and system based on cross-modal multi-scale feature guidance

The invention discloses a depth map super-resolution method and system based on cross-modal multi-scale feature guidance, and the method comprises the steps: respectively inputting a low-resolution depth map and a corresponding high-resolution color image in the same scene into a feature extraction branch, extracting the features of a tth stage, and obtaining a group of depth map features and a group of color map features; inputting the depth map feature and the color map feature at the moment t = 0 into a feature fusion branch to obtain an initial fusion feature; sending the initial fusion feature and the depth map feature and the color map feature at the current moment into a feature fusion branch to obtain a fusion feature at the moment t = 1, and repeating for multiple times to obtain a fusion feature at the moment t; the fusion features at the t moment are sent to a depth image super-resolution reconstruction branch to obtain a high-resolution residual image; and magnifying the original low-resolution de