keras实现残差网络（keras搬砖二）

 1 from keras import layers
 2 import keras
 3 import numpy as np
 4 
 5 inputs = keras.Input(shape=(32, 32, 3), name="img")
 6 x = layers.Conv2D(32, 3, activation="relu")(inputs)
 7 x = layers.Conv2D(64, 3, activation="relu")(x)
 8 block_1_output = layers.MaxPooling2D(3)(x)
 9 
10 x = layers.Conv2D(64, 3, activation="relu", padding="same")(block_1_output)
11 x = layers.Conv2D(64, 3, activation="relu", padding="same")(x)
12 block_2_output = layers.add([x, block_1_output])
13 
14 x = layers.Conv2D(64, 3, activation="relu", padding="same")(block_2_output)
15 x = layers.Conv2D(64, 3, activation="relu", padding="same")(x)
16 block_3_output = layers.add([x, block_2_output])
17 
18 x = layers.Conv2D(64, 3, activation="relu")(block_3_output)
19 x = layers.GlobalAveragePooling2D()(x)
20 x = layers.Dense(256, activation="relu")(x)
21 x = layers.Dropout(0.5)(x)
22 outputs = layers.Dense(10)(x)
23 
24 model = keras.Model(inputs, outputs, name="toy_resnet")
25 model.summary()
26 
27 # 绘制模型
28 keras.utils.plot_model(model, "mini_resnet.png", show_shapes=True)
29 
30 # 训练模型
31 (x_train, y_train), (x_test, y_test) = keras.datasets.cifar10.load_data()
32 
33 x_train = x_train.astype("float32") / 255.0
34 x_test = x_test.astype("float32") / 255.0
35 y_train = keras.utils.to_categorical(y_train, 10)
36 y_test = keras.utils.to_categorical(y_test, 10)
37 
38 model.compile(
39     optimizer=keras.optimizers.RMSprop(1e-3),
40     loss="categorical_crossentropy",
41     metrics=["accuracy"],
42 )
43 # We restrict the data to the first 1000 samples so as to limit execution time
44 # on Colab. Try to train on the entire dataset until convergence!
45 model.fit(x_train[:1000], y_train[:1000], batch_size=64, epochs=1, validation_split=0.2)