reviseIn addition in each stage of encoder the channels of all feature maps are reduced to 128 before inputting the FPE module In FPE module the first two blocks adopt 5×5 and 7×7 average pooling ope

Additionally, before inputting the FPE module, the channels of all feature maps are reduced to 128 in each stage of the encoder. Within the FPE module, the first two blocks utilize 5×5 and 7×7 average pooling operations, while the last two blocks employ 3×3 and 5×5 pooling kernels. It is worth noting that the sizes of all edge feature maps are made equal through the use of padding operations. Finally, the output of the decoder is directly bilinear interpolated to match the size of the input image