Fig. 1. MsiNet structure

Fig. 2. Structures of RCU and IRU and loss function comparison. (a) General network structure of residual convolution unit (RCU) in ResNet; (b) architecture of interlaced residual unit (IRU) in MsiNet; (c) loss comparison chart during training

Fig. 3. Structures of each fusion unit. (a) Structure of biased fusion unit (BFU) in MsiNet; (b) architecture of fusion unit in RefineNet; (c) CFF structure in ICNet

Fig. 4. Original OCT image (a), Ground Truth image (b), and generalized semantic boundary(c)

Fig. 5. Variation of accuracy with θ on GSE

Fig. 6. Representative qualitative comparison for retina segmentation when different values of θ are applied

Fig. 7. Comparison of three outputs in MsiNet

Fig. 8. Qualitative comparison of layer segmentation