Fig. 11:

Examples of feature maps in TransMorph’s skip connections. Eight feature maps are randomly selected from the feature maps associated with each skip connection. Left panel: Example 2D slices of source and target images (i.e., $I_m$ and $I_f$), which are used as inputs to TransMorph. Middle panel: Feature maps in the skip connections of the two convolutional layers (denoted by the green arrows in Fig. 1). Right panel: Feature maps in the skip connections of the Swin Transformer blocks (denoted by the orange arrows in Fig. 1).