我正在使用预训练的Densenet网络进行皮肤病变检测和识别工作。然而，我遇到了一个问题，提示说权重无法广播到值。
```%tensorflow_version 1.ximport tensorflow as tffrom tensorflow.keras.applications import densenetfrom tensorflow.keras import layersfrom tensorflow.keras.models import Sequential,Model,load_model from tensorflow.keras.layers import Convolution2Dfrom tensorflow.keras.layers import MaxPooling2Dfrom tensorflow.keras.layers import Flatten,Dense,Activation,Dropoutfrom tensorflow.keras.callbacks import EarlyStopping,ReduceLROnPlateau,ModelCheckpoint,Callbackfrom tensorflow.keras.preprocessing.image import ImageDataGeneratorfrom tensorflow.keras.models import Model image_width,image_height=224,224 #分配高度和宽度 --- Densenet在224*224尺寸上训练 training_samples = 33068 #训练样本数testing_samples = 1103 #验证样本数epochs = 100batch_size = 10 #批次大小设为10 n_classes = 7 #类别数为7。我们有7种皮肤病类别training_folder = '/content/gdrive/My Drive/Deep Learning Projects/skincancer/skin_cancer_classified/training_directory/'testing_folder = '/content/gdrive/My Drive/Deep Learning Projects/skincancer/skin_cancer_classified/testing_directory/'datagen = ImageDataGenerator( preprocessing_function= \ tf.keras.applications.densenet.preprocess_input)#使用与densenet架构原始rgb图像相同的预处理技术training_batches = datagen.flow_from_directory(training_folder, target_size=(image_height,image_width), batch_size = batch_size)testing_batches = datagen.flow_from_directory(testing_folder, target_size=(image_height,image_width), batch_size = batch_size)network = tf.keras.applications.densenet.DenseNet121()######通过移除网络的最后5层来创建模型架构engineered_network = network.layers[-5].outputengineered_network = Dropout(0.25)(engineered_network)predictions = Dense(7,activation='softmax')(engineered_network) final_model = Model(inputs = network.input, outputs = predictions)for layer in final_model.layers[:-30]: layer.trainable = Falsefinal_model.compile(optimizer='Adam',loss='categorical_crossentropy',metrics=["accuracy"])class_weights = { 0: 1.0, 1: 1.0, 2: 1.0, 3: 1.0, 4: 3.2, #使黑色素瘤更敏感 5: 1.0, 6: 1.0,}network_training = final_model.fit_generator(training_batches, steps_per_epoch=36, class_weight=class_weights, validation_data = testing_batches, validation_steps = 110, epochs=epochs, verbose = 1)```我不断收到这个错误：
#############################################Epoch 1/100---------------------------------------------------------------------------ValueError Traceback (most recent call last)<ipython-input-16-6a6e3a644f6a> in <module>() 5 validation_steps = 110, 6 epochs=epochs,----> 7 verbose = 1)11 frames/tensorflow-1.15.2/python3.6/tensorflow_core/python/ops/weights_broadcast_ops.py in assert_broadcastable(weights, values) 101 " values.shape=%s. weights.shape=%s." % ( 102 _ASSERT_BROADCASTABLE_ERROR_PREFIX, values_rank_static,--> 103 weights_rank_static, values.shape, weights.shape)) 104 weights_shape_static = tensor_util.constant_value(weights_shape) 105 values_shape_static = tensor_util.constant_value(values_shape)ValueError: 权重无法广播到值。值的秩为3，权重的秩为1。值的形状为(?, 7, 7)，权重的形状为(?,)。我之前没有使用过预训练网络，所以这是我第一次遇到这种情况。任何帮助都将不胜感激。谢谢
回答：
我认为您可能试图直接在卷积层上添加一个全连接层。尝试将预训练模型的输出展平：
######通过移除网络的最后5层来创建模型架构engineered_network = network.layers[-5].outputengineered_network = Dropout(0.25)(engineered_network)engineered_network = Flatten()(engineered_network)predictions = Dense(7,activation='softmax')(engineered_network) 