ArtNet.py 文件源码

def loss_style(self, gen_img_rep):
        feat_cor_gen = self.feature_cor(gen_img_rep)

        feat_loss = 0
        for i in range(len(feat_cor_gen)):
            orig_shape = self.style_rep[i].shape
            feat_map_size = orig_shape[2] * orig_shape[3]  # M_l

            layer_wt = 4.0 * feat_map_size ** 2.0

            feat_loss += F.mean_squared_error(self.style_feat_cor[i], feat_cor_gen[i]) / layer_wt

        return feat_loss

    # total loss function
    # cf. equation (7) of the article