我将以下密集层作为我的最后一层。
flat = Flatten()(reduction_pose)
pose = Dense(1024, activation='relu', name='pose')(flat)
xyz = Dense(2, name='xyz')(pose)
model_out = Model(input=input, output=[xyz])
我想用完全连接的层替换密集层。我该怎么办?我尝试了以下方法,
conv10 = Convolution2D(2, (1, 1), name='conv_10_1')(reduction_pose)
x = Reshape((224 * 224, 2))(conv10)
x = Activation('softmax')(x)
outputs = Reshape((224, 224, 2))(x)
model_out = Model(inputs=input, outputs=outputs)
但是我遇到了以下错误,
ValueError:新数组的总大小必须保持不变
错误指向x = Reshape((224 * 224, 2))(conv10)