Fig. 3
From: Are RGB-based salient object detection methods unsuitable for light field data?
Network architecture of the proposed encoder and predictor. Each convolutional block consists of a convolutional layer, a batch normalization layer, and a ReLU nonlinearity layer