Fig. 4 | Scientific Reports

Fig. 4

From: Deep multi-task learning framework for gastrointestinal lesion-aided diagnosis and severity estimation

Fig. 4

Experimental sample multi-head attention maps (8 heads). Visualization of the lesion diagnosis results of the proposed method. The 1st column represents the input (original), and the 2nd column to the 8th column represents the generated attention head, respectively. As we can see, the proposed model can accurately localize and detect lesions on GT regardless of their size.

Back to article page