Table 1.

Results on Retouch dataset. Model size (M) and testing speed (FPS) are also reported. The best results are bold.

Method	$\mathit{D}\mathit{S}{\mathit{C}}_{\mathit{v}}$	$\mathit{D}\mathit{S}\mathit{C}$	$\mathit{A}\mathit{V}\mathit{D}$	$\mathit{B}\mathit{A}\mathit{C}\mathit{C}$	Size / Speed
DeepLabv3+ [57]	60.2	82.7	0.023	86.5	59.3 / 38
U-Net [58]	66.1	84.2	0.021	87.4	13.4 / 38
Att-UNet [59]	65.3	83.4	0.022	86.6	34.9 / 36
CE-Net [60]	67.3	84.2	0.026	84.6	29.0 / 37
nnU-Net [61]	67.2	84.3	0.023	86.4	30.0 / 20
CPFNet [62]	69.0	85.7	0.022	88.0	43.3 / 37
Curvature [63]	68.2	84.7	0.024	87.1	43.3 / 37
MsTGANet [64]	68.9	85.0	0.023	87.1	11.6 / 37
DconnNet	78.2	87.7	0.020	90.5	36.4 / 40