TABLE I.

Architecture of the proposed network. All convolutional layers include the dropout and batch normalization. Note that “3×3|64” corresponds to 3×3 kernel generating 64 features and tconv means the transposed convolution.

Stage	Encoding part	Object-dependent up sampling	Skip connection	Decoding part	Fully connected
1	$\left[\begin{array}{cc}3\times 3\mid 64& conv,ReLU\\ 3\times 3\mid 64& conv,ReLU\end{array}\right]$ $\left[\begin{array}{cc}2\times 2& \text{max}\mathit{pool}\end{array}\right]$	$[\begin{array}{cc}3\times 3\mid 64& tconv,ReLU\end{array}$	$[\begin{array}{cc}3\times 3\mid 64& conv,ReLU\end{array}$	$\left[\begin{array}{cc}2\times 2& \mathit{deconv}\end{array}\right]$ $\left[\begin{array}{cc}3\times 3\mid 64& conv,ReLU\\ 3\times 3\mid 64& conv,ReLU\end{array}\right]$	$\left[\begin{array}{cc}1\times 1& \mathit{conv}\end{array}\right]$
2	$\left[\begin{array}{cc}3\times 3\mid 128& conv,ReLU\\ 3\times 3\mid 128& conv,ReLU\end{array}\right]$ $\left[\begin{array}{cc}2\times 2& \text{max}\mathit{pool}\end{array}\right]$	$[\begin{array}{cc}3\times 3\mid 128& tconv,ReLU\end{array}$	$[\begin{array}{cc}3\times 3\mid 128& conv,ReLU\end{array}$	$\left[\begin{array}{cc}2\times 2& \mathit{deconv}\end{array}\right]$ $\left[\begin{array}{cc}3\times 3\mid 128& conv,ReLU\\ 3\times 3\mid 128& conv,ReLU\end{array}\right]$	-
3	$\left[\begin{array}{cc}3\times 3\mid 256& conv,ReLU\\ 3\times 3\mid 256& conv,ReLU\end{array}\right]$ $\left[\begin{array}{cc}2\times 2& \text{max}\mathit{pool}\end{array}\right]$	$[\begin{array}{cc}3\times 3\mid 256& tconv,ReLU\end{array}$	$[\begin{array}{cc}3\times 3\mid 256& conv,ReLU\end{array}$	$\left[\begin{array}{cc}2\times 2& \mathit{deconv}\end{array}\right]$ $\left[\begin{array}{cc}3\times 3\mid 256& conv,ReLU\\ 3\times 3\mid 256& conv,ReLU\end{array}\right]$	-
4	$\left[\begin{array}{cc}3\times 3\mid 512& conv,ReLU\\ 3\times 3\mid 512& conv,ReLU\end{array}\right]$ $\left[\begin{array}{cc}2\times 2& \text{max}\mathit{pool}\end{array}\right]$	$[\begin{array}{cc}3\times 3\mid 512& tconv,ReLU\end{array}$	$[\begin{array}{cc}3\times 3\mid 512& conv,ReLU\end{array}$	$\left[\begin{array}{cc}2\times 2& \mathit{deconv}\end{array}\right]$ $\left[\begin{array}{cc}3\times 3\mid 512& conv,ReLU\\ 3\times 3\mid 512& conv,ReLU\end{array}\right]$	-
5	$\left[\begin{array}{cc}3\times 3\mid 1024& conv,ReLU\\ 3\times 3\mid 1024& conv,ReLU\end{array}\right]$	-	-	-	-