Skip to main content
. Author manuscript; available in PMC: 2019 Apr 11.
Published in final edited form as: Med Image Comput Comput Assist Interv. 2018 Sep 26;11070:871–879. doi: 10.1007/978-3-030-00928-1_98

Fig. 1.

Fig. 1

Architecture of the multi-task SonoEyeNet (M-SEN). It has two modules: the generator (in Green-dashed polygon) and the discriminator (Orange-dashed box). The generator has two tasks: a primary task to classify frames (bottom) and an auxiliary task to predict visual attention map (Â). The discriminator differentiates between real (A) and predicted (Â) attention maps. The dotted circle ⊙ indicates element-wise multiplication. LS, LC and LD represent the losses of saliency prediction, frame classfication, and the discriminator respectively.