Fig. 3.
The network architecture of Attention Residual Dense U-Net (AttRDUNet) used for the image restoration network Gimg and sinogram restoration network Gsino in DuDoDR-Net (Figure 2). Our AttRDUNet consists of Attention Residual Dense Blocks (AttRDB) at different resolution levels, and convolutional GRU (convGRU) in the bottleneck for recurrent learning.