Multitask Interactive Attention Learning Model Based on Hand Images for Assisting Chinese Medicine in Predicting Myocardial Infarction

Figure 4

The framework of the proposed IIM. When the attention modules (task M and task P) of different tasks are input into the IIM, their feature maps are regarded as the main information and reference information, respectively. After the reference information is given a certain weight (blue translucent cube and orange translucent cube), it interacts with the main information. The useful channels of the final feature map are emphasized according to the depth of channel color.