A grayscale image automatic coloring method based on a multi-view attention mechanism

The invention relates to a grayscale image automatic coloring method based on a multi-view attention mechanism, and belongs to the technical field of grayscale image coloring. The method comprises the following steps: 1, collecting and processing data; 2, constructing a model, carrying out global average pooling on the feature map F, carrying out channel pooling on the feature map F, and converting the input feature map F into Fout_3 without any transformation. The invention has the beneficial effects that a multi-view attention mechanism is adopted, and the multi-view attention module can enable the coloring model to adaptively obtain attention information of different views and different scales, so that the model is better supported to reconstruct a color image. The feature information of different view angles can be adaptively learned, the feature extraction capability of a coding-decoding structure is improved, and modeling of semantic information and context information in an image is enhanced. The modul