python ValueError：必须始终传递ウLayer.call Ж的第一个参数,我希望以子类化的方式键入此模型,

von4xj4u  于 2022-10-30  发布在  Python
关注(0)|答案(1)|浏览(164)

我有一些不同的块，因为我必须使用共享参数，所以我需要以子类形式实现此模型

卷积和恒等块

  1. def convolutional_block(x, filter):
  2.     # copy tensor to variable called x_skip
  3.     x_skip = x
  4.     # Layer 1
  5.     x = tf.keras.layers.Conv2D(filter, (1, 1), strides = (2, 2))(x)
  6.     x = tf.keras.layers.BatchNormalization(axis=3)(x)
  7.     x = tf.keras.layers.Activation('relu')(x)
  8.     # Layer 2
  9.     x = tf.keras.layers.Conv2D(filter, (3, 3), padding = 'same')(x)
  10.     x = tf.keras.layers.BatchNormalization(axis=3)(x)
  11.     x = tf.keras.layers.Activation('relu')(x)
  12.     # Layer 3
  13.     x = tf.keras.layers.Conv2D(filter * 4, (1, 1))(x)
  14.     x = tf.keras.layers.BatchNormalization(axis=3)(x)
  15.     # Processing Residue with conv(1,1)
  16.     x_skip = tf.keras.layers.Conv2D(filter * 4, (1, 1), strides = (2, 2))(x_skip)
  17.     # Add Residue
  18.     x = tf.keras.layers.Add()([x, x_skip])
  19.     x = tf.keras.layers.Activation('relu')(x)
  20.     return x
  22. def identity_block(x, filter, name):
  23.     # copy tensor to variable called x_skip
  24.     x_skip = x
  25.     # Layer 1
  26.     x = tf.keras.layers.Conv2D(filter, (1, 1))(x)
  27.     x = tf.keras.layers.BatchNormalization(axis=3)(x)
  28.     x = tf.keras.layers.Activation('relu')(x)
  29.     # Layer 2
  30.     x = tf.keras.layers.Conv2D(filter, (3, 3), padding = 'same')(x)
  31.     x = tf.keras.layers.BatchNormalization(axis=3)(x)
  32.     x = tf.keras.layers.Activation('relu')(x)
  33.     # Layer 3
  34.     x = tf.keras.layers.Conv2D(filter * 4, (1, 1))(x)
  35.     x = tf.keras.layers.BatchNormalization(axis=3)(x)
  36.     # Add Residue
  37.     x = tf.keras.layers.Add()([x, x_skip])
  38.     x = tf.keras.layers.Activation('relu', name=f'identity_{name}')(x)
  39.     return x

互斥锁块

  1. class MutexAttentionBlock(tf.keras.layers.Layer):
  2.     def __init__(self):
  3.         super(MutexAttentionBlock, self).__init__()
  4.         self.softmax = tf.keras.layers.Activation('softmax')
  5.         self.multiply = tf.keras.layers.Multiply()
  7.     def call(self, x, y):
  8.         distance = tf.math.square(tf.math.subtract(x, y))
  9.         x = tf.keras.layers.Reshape((-1, distance.shape[-1]))(distance)
  10.         x = self.softmax(x)
  11.         x = tf.keras.layers.Reshape((distance[1], distance[2], distance[3]))(x)
  12.         f_am = self.multiply(x, y)
  13.         return f_am

融合块

  1. class FuseAttentionBlock(tf.keras.layers.Layer):
  2.     def __init__(self):
  3.         super(FuseAttentionBlock, self).__init__()
  4.         self.first_add = tf.keras.layers.Add()
  5.         self.global_avg_pool = tf.keras.layers.GlobalAveragePooling2D()
  6.         self.global_max_pool = tf.keras.layers.GlobalMaxPooling2D()
  7.         self.second_add = tf.keras.layers.Add()
  8.         self.concat = tf.keras.layers.Concatenate(axis=0)
  9.         self.softmax = tf.keras.layers.Softmax(axis=0)
  10.         self.first_multiply = tf.keras.layers.Multiply()
  11.         self.second_multiply = tf.keras.layers.Multiply()
  12.         self.third_add = tf.keras.layers.Add()
  14.     def call(self, f_am, y):
  15.         f_mix = self.first_add([f_am, y])
  17.         x = self.global_avg_pool(f_mix)
  18.         y = self.global_max_pool(f_mix)
  20.         x = self.second_add([x, y])
  22.         for i in range(2):
  23.             x = tf.keras.layers.Dense(32)(x)
  24.             x = tf.keras.layers.BatchNormalization()(x)
  25.             x = tf.keras.layers.Activation('relu')(x)
  27.         m = tf.keras.layers.Dense(f_am.shape[-1])(x)
  28.         n = tf.keras.layers.Dense(y.shape[-1])(x)
  30.         concated = self.concat([m, n], axis=0)
  32.         attention_weights = self.softmax(concated)
  34.         f_am = self.first_multiply([f_am, attention_weights])
  35.         y = self.second_multiply([y, attention_weights])
  37.         f_fm = self.third_add([f_am, y])
  38.         return f_fm

最终架构

  1. class MutexAttentionResModel(tf.keras.models.Model):
  2.     def __init__(self, shape, num_classes):
  3.         super(MutexAttentionResModel, self).__init__()
  4.         self.shape=shape
  5.         self.num_classes = num_classes
  7.         input_img = tf.keras.layers.Input(self.shape)
  8.         x = tf.keras.layers.ZeroPadding2D((3, 3))(input_img)
  9.         x = tf.keras.layers.Conv2D(64, kernel_size=7, strides=2, padding='same')(x)
  10.         x = tf.keras.layers.BatchNormalization()(x)
  11.         x = tf.keras.layers.Activation('relu')(x)
  12.         x = tf.keras.layers.MaxPool2D(pool_size=3, strides=2, padding='same')(x)
  13.         self.first_layer = tf.keras.Model(inputs=input_img, outputs=x)
  15.         input_second_layer = self.first_layer.output
  16.         x = convolutional_block(input_second_layer, 64)
  17.         for i in range(2):
  18.             x = identity_block(x, 64, name=f'2_{i}')
  19.         self.second_layer = tf.keras.Model(inputs=input_second_layer, outputs=x)
  21.         input_third_layer = self.second_layer.output
  22.         x = convolutional_block(input_third_layer, 128)
  23.         for i in range(3):
  24.             x = identity_block(x, 128, name=f'3_{i}')
  25.         self.third_layer = tf.keras.Model(inputs=input_third_layer, outputs=x)
  27.         input_fourth_layer = self.third_layer.output
  28.         x = convolutional_block(input_fourth_layer, 256)
  29.         for i in range(5):
  30.             x = identity_block(x, 256, name=f'4_{i}')
  31.         self.fourth_layer = tf.keras.Model(inputs=input_fourth_layer, outputs=x)
  33.         input_fifth_layer = self.fourth_layer.output
  34.         x = convolutional_block(input_fifth_layer, 512)
  35.         for i in range(2):
  36.             x = identity_block(x, 512, name=f'5_{i}')
  37.         self.fifth_layer = tf.keras.Model(inputs=input_fifth_layer, outputs=x)
  39.         self.first_mutex_attention_block = MutexAttentionBlock()
  40.         self.first_fuse_attention_block = FuseAttentionBlock()
  42.         self.second_mutex_attention_block = MutexAttentionBlock()
  43.         self.second_fuse_attention_block = FuseAttentionBlock()
  45.         self.third_mutex_attention_block = MutexAttentionBlock()
  46.         self.third_fuse_attention_block = FuseAttentionBlock()
  48.         self.fourth_mutex_attention_block = MutexAttentionBlock()
  49.         self.fourth_fuse_attention_block = FuseAttentionBlock()
  51.         self.fifth_mutex_attention_block = MutexAttentionBlock()
  52.         self.fifth_fuse_attention_block = FuseAttentionBlock()
  54.         self.global_max_pool = tf.keras.layers.GlobalMaxPooling2D()
  56.         self.first_dense = tf.keras.layers.Dense(self.num_classes, activation='softmax')
  57.         self.second_dense = tf.keras.layers.Dense(self.num_classes, activation='softmax')
  59.     def call(self, pos_input, neg_input):
  60.         x = self.first_layer(pos_input)
  61.         y = self.first_layer(neg_input)
  62.         f_am = self.first_mutex_attention_block(x, y)
  63.         y = self.first_fuse_attention_block(f_am, y)
  65.         x = self.second_layer(x)
  66.         y = self.second_layer(y)
  67.         f_am = self.second_mutex_attention_block(x, y)
  68.         y = self.second_fuse_attention_block(f_am, y)
  70.         x = self.third_layer(x)
  71.         y = self.third_layer(y)
  72.         f_am = self.third_mutex_attention_block(x, y)
  73.         y = self.third_fuse_attention_block(f_am, y)
  75.         x = self.fourth_layer(x)
  76.         y = self.fourth_layer(y)
  77.         f_am = self.fourth_mutex_attention_block(x, y)
  78.         y = self.fourth_fuse_attention_block(f_am, y)
  80.         x = self.fifth_layer(x)
  81.         y = self.fifth_layer(y)
  82.         f_am = self.fifth_mutex_attention_block(x, y)
  83.         y = self.fifth_fuse_attention_block(f_am, y)
  85.         x_flatten = tf.keras.layers.Flatten(name='x_flatten')(x)
  86.         y_flatten = tf.keras.layers.Flatten(name='y_flatten')(y)
  88.         x = self.global_max_pool(x)
  89.         y = self.global_max_pool(y)
  91.         x = self.first_dense(x)
  92.         y = self.second_dense(y)
  94.         return x, y, x_flatten, y_flatten

尝试取得子类别模型的摘要

  1. model = MutexAttentionResModel(shape = config.IMG_SHAPE, num_classes = 2)
  3. image = plt.imread('architecture\images.jpg')
  4. image = tf.image.resize(image,[224,224])
  5. image = tf.cast(image,dtype = tf.float32)
  7. image = tf.expand_dims(image,0)
  8. output = model(ct_input= image,mutex_input = image)
  9. print(image.shape)
  10. model.summary()

我接受这个错误我不知道该怎么办.

值错误：必须始终传递Layer.call的第一个参数。

python

1条答案

按热度按时间
jchrr9hc

jchrr9hc1#

试试这个，但我会告诉你这个错误背后的原因，这个错误发生时，你使用相同的两个层的一个名称，并将其发送到keras.模型，我已经改变了代码一点点，并不返回模型输出在每一层只是返回x它是一样的，因为你正在做的，但复杂性很低...
第一个问题：

赞(0）分享  回复(0）举报  2022-10-30
我来回答

相关问题

    查看更多

    热门标签

    更多
    JavaquerypythonNode开发语言requestUtil数据库Table后端算法LoggerMessageElementParser

    最新问答

    更多
    • 蜀ICP备13028337号-1 大数据知识库 https://www.saoniuhuo.com © All rights reserved
    • 本站内容来源互联网,如果侵犯您的权益请联系我们删除, 联系方式：448109455@qq.com