Mask image modeling algorithm based on self-supervised learning

The invention relates to the technical field of MIM (Mask Image Modeling), in particular to a mask image modeling algorithm based on Self-Supervised Learning), which comprises the following steps: firstly, dividing an image into patches and randomly dividing the patches into four equal parts, taking each part as a visible patch and the rest as mask patches so as to obtain four mask images, taking the visible patches as encoder input to obtain potential feature representation, and taking the visible patches as encoder input to obtain the mask image; and the coded visible path and mask path are jointly used as the input of a decoder for image reconstruction, and the certainty of a model reconstruction result is enhanced by minimizing the mean absolute error of the prediction results of the overlapped parts of the mask paths in different mask images obtained from the same image. According to the method, data are fully utilized, training time and hardware resources are greatly saved, and the method is located at