... the extracted features, the author develops a text-guided multi-head attention fusion mechanism module. Subsequently, a directed graph is constructed through SE and the multimodal factorized bilinear pooling method to integrate image features into the graph. The DGIN then fuses the image and text features...