the loss function of each cycle is

the loss function of each cycle is defined as the mean square error(MSE) between the attention map output At at the tth cycle and mask M, which is the result of subtracting the background image from the input rainy image. The formula is written as follows,Where At is the attention map generated from the tth attention cycle, At = ATTt(Ft-1, Ht-1, Ct-1), Ft-1 is connection of the input rainy image and the attention map generated from the last cycle, when t=1, which is at the process of the first loop block in the recurrent network, the attention map is set as 0.5, more cycles would produce better attention maps, but it is also computer memory required. Set the time steps =4, ϴ = 0.8.