Histopathological Tissue Segmentation of Lung Cancer with Bilinear CNN and Soft Attention

Figure 3

An overview of the proposed classification network. First, the images are fed into ResNet50 to get the feature maps, and feature maps are input to the bilinear function to obtain the bilinear vector; then, the attention weight is got from the attention function, and finally, the bilinear vector multiplies the attention weight to obtain the attention feature flowed into the softmax layer for classification.