lostcorner

> @kk-kaii 我的理解是已经做了的。 ``` x4_3_up = F.upsample(x4_3, size=x3_2.size()[2:], mode='bilinear') x4_3_up_map1 = self.conv_3(x4_3_up) x3_2_map1 = self.conv_3(x3_2) x4_3_up_map2 = self.conv_5(x4_3_up) x3_2_map2 = self.conv_5(x3_2) x4_3_2 = self.x4_x3_x2(abs(x4_3_up - x3_2)+abs(x4_3_up_map1-x3_2_map1)+abs(x4_3_up_map2-x3_2_map2) ) ``` 以这一段代码为例，map1和map2分别代表用conv3和conv5卷积后的结果。在最后的相加过程中分别有三次绝对值，对应1x1, 3x3,...