Depth completion method based on distance perception mask converter and sparse attention mask mechanism

The invention belongs to the field of deep learning, computer vision, image processing and image super-resolution reconstruction, and discloses a depth completion method based on a distance perception mask converter and a sparse attention mask mechanism so as to better realize a depth completion task. The whole method is based on an innovative macro-to-micro network, and specific details in sparse depth are enhanced in a micro range while priori knowledge in a macro data set range is utilized. The performance of reconstructing a dense and detailed depth map from a sparse depth map paired with a corresponding image is greatly improved. According to the method, a distance perception mask converter decoder is designed, and an enhanced context about missing values in a sparse depth map is provided by using prior information contained in a data set; a sparse attention mask is designed, an invalid region is shielded, and the performance is automatically improved, so that rich texture information in the RGB image is