我想根据本文创建损失函数（http://mediatum.ub.tum.de/doc/1463390/488043490645.pdf）
为此，我必须根据该论文添加方差层来更改当前的SSD对象检测器
我的问题是我们如何在ssd模型上创建方差层
我们如何为盒子和类创建损失函数
我尝试如下在ssd上创建图层
但我不确定是否与论文中所述
相同classes4_reshaped = Reshape((-1, num_classes), name='classes4_reshape')(classes4)
#We added dropout layer because yarian gal said that dropout layer should added before output layer & on top of fully connected layer
classes4_reshaped = Dropout(0.5)(classes4_reshaped)
classes5_reshaped = Reshape((-1, num_classes), name='classes5_reshape')(classes5)
classes5_reshaped = Dropout(0.5)(classes5_reshaped)
classes6_reshaped = Reshape((-1, num_classes), name='classes6_reshape')(classes6)
classes6_reshaped = Dropout(0.5)(classes6_reshaped)
classes7_reshaped = Reshape((-1, num_classes), name='classes7_reshape')(classes7)
classes7_reshaped = Dropout(0.5)(classes7_reshaped)
# Reshape the box coordinate predictions, yielding 3D tensors of shape `(batch, height * width * n_boxes, 4)`
# We want the four box coordinates isolated in the last axis to compute the smooth L1 loss
boxes4_reshaped = Reshape((-1, 4), name='boxes4_reshape')(boxes4)
boxes4_reshaped = Dropout(0.5)(boxes4_reshaped)
boxes5_reshaped = Reshape((-1, 4), name='boxes5_reshape')(boxes5)
boxes5_reshaped = Dropout(0.5)(boxes5_reshaped)
boxes6_reshaped = Reshape((-1, 4), name='boxes6_reshape')(boxes6)
boxes6_reshaped = Dropout(0.5)(boxes6_reshaped)
boxes7_reshaped = Reshape((-1, 4), name='boxes7_reshape')(boxes7)
boxes7_reshaped = Dropout(0.5)(boxes7_reshaped)
anchors4_reshaped = Reshape((-1, 8), name='anchors4_reshape')(anchors_4)
anchors5_reshaped = Reshape((-1, 8), name='anchors5_reshape')(anchors_5)
anchors6_reshaped = Reshape((-1, 8), name='anchors6_reshape')(anchors_6)
anchors7_reshaped = Reshape((-1, 8), name='anchors7_reshape')(anchors_7)
variance_pre =密集（1，name ='variance_pre'）（classes_concat） 方差=激活（'softplus'，名称='方差'）（variance_pre）
logits_variance_class = Concatenate(axis=2,name='logits_variance_class')([classes_concat, variance])
boxes_concat = Concatenate(axis=1, name='boxes_concat')([boxes4_reshaped,
boxes5_reshaped,
boxes6_reshaped,
boxes7_reshaped])
variance_pre_box = Dense(1, name='variance_pre_box')(boxes_concat)
variance_box = Activation('softplus', name='variance_box')(variance_pre_box)
logits_variance_box =串联（轴= 2，名称='logits_variance_box'）（[boxes_concat，variance_box]）
logits_variance_class =并置（axis = 2，name ='logits_variance_class'）（[[classes_concat，variation]）
logits_variance =串联（axis = 2，name ='logits_variance'）（[logits_variance_class，logits_variance_box，anchors_concat]）
model2 =模型（输入= x，输出= logits_variance）
我需要一些建议，根据本文我们必须如何执行模型和损失方程。