Figure 2.

ESFPNet decoder modules. Inputs and outputs are given by the top and bottom feature tensor blocks, with dimensions as indicated. For the $\mathbf{ConvModule}(·)$, dimension $C_i$ arises from stage output $F_i$ of the MiT encoder. (a) $\mathbf{Linear}\_\mathbf{Layer}(·)$. (b) $\mathbf{ConvModule}(·)$. (c) $\mathbf{Multistage}\_\mathbf{Fusion}(·)$.