ResNet如何實現

今天我們要來實作ResNet，並比較加入ResNet後，是否正確率有提升。

前置作業

我們使用Colab來當作我們的實作平台，並使用Keras來完成。

資料集

fashion_mnist，為Keras內建的資料集
訓練集為60,000 張28x28 像素灰度圖像，測試集為10,000 同規格圖像，總共10 類時尚物品標籤。該數據集可以用作MNIST 的直接替代品。類別標籤是：
類別 描述 中文
0 T-shirt/top T卹/上衣
1 Trouser 褲子
2 Pullover 套頭衫
3 Dress 連衣裙
4 Coat 外套
5 Sandal 涼鞋
6 Shirt 襯衫
7 Sneaker 運動鞋
8 Bag 背包
9 Ankle boot 短靴

Code

前置處理

讀取資料集後，把數值都scale到0~1之間。
並設定batch_size=150、epoch=25

from keras.layers import Input, Dense, Conv1D, Conv2D, MaxPooling1D,\
    MaxPooling2D, UpSampling1D, UpSampling2D, Dropout, Lambda, Convolution2D,\
    Reshape, Activation, Flatten, add, concatenate, Subtract, BatchNormalization
from keras.models import Model, Sequential
from keras.datasets import fashion_mnist
import numpy as np
import keras

nb_classes=10
nb_epoch=25
batch_size=150

(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()
x_train = x_train.astype('float32') / 255.0
x_test = x_test.astype('float32') / 255.0
x_train = np.reshape(x_train, (x_train.shape[0], x_train.shape[1], x_train.shape[2], 1))
x_test = np.reshape(x_test, (x_test.shape[0], x_test.shape[1], x_test.shape[2], 1))

model

12層

#12層
tag = 'ResNet0910_fashion_mnist_12layer_{}'.format(0)
h5_weight_path = os.path.join(WEIGHT_DIR, './' + tag + '.h5')

input_shape=(28,28,1)

input = Input(input_shape, name='input')

layer=Conv2D(32, kernel_size=(2, 2), activation='relu', padding='same')(input)
layer=BatchNormalization()(layer)
for i in range(2):
  layer=Conv2D(32, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=MaxPooling2D(pool_size=(2, 2))(layer)
for i in range(3):
  layer=Conv2D(64, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=MaxPooling2D(pool_size=(2, 2))(layer)
for i in range(3):
  layer=Conv2D(128, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=MaxPooling2D(pool_size=(2, 2))(layer)
for i in range(3):
  layer=Conv2D(256, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=MaxPooling2D(pool_size=(2, 2))(layer)

layer = Dropout(0.5)(layer)
layer = Flatten(name='flatten')(layer)
output = Dense(nb_classes, name="Dense_10nb", activation='softmax')(layer)

model = Model(inputs=[input], outputs=[output])

model.compile(loss='sparse_categorical_crossentropy',optimizer=keras.optimizers.Adam(lr=0.0001,decay=1e-6),metrics = ['accuracy'])
model.summary()

20層

#20層
tag = 'ResNet0910_fashion_mnist_20layer_{}'.format(0)
h5_weight_path = os.path.join(WEIGHT_DIR, './' + tag + '.h5')

input_shape=(28,28,1)

input = Input(input_shape, name='input')

layer=Conv2D(32, kernel_size=(2, 2), activation='relu', padding='same')(input)
layer=BatchNormalization()(layer)
for i in range(4):
  layer=Conv2D(32, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=MaxPooling2D(pool_size=(2, 2))(layer)
for i in range(5):
  layer=Conv2D(64, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=MaxPooling2D(pool_size=(2, 2))(layer)
for i in range(5):
  layer=Conv2D(128, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=MaxPooling2D(pool_size=(2, 2))(layer)
for i in range(5):
  layer=Conv2D(256, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=MaxPooling2D(pool_size=(2, 2))(layer)

layer = Dropout(0.5)(layer)
layer = Flatten(name='flatten')(layer)
output = Dense(nb_classes, name="Dense_10nb", activation='softmax')(layer)

model = Model(inputs=[input], outputs=[output])

model.compile(loss='sparse_categorical_crossentropy',optimizer=keras.optimizers.Adam(lr=0.0001,decay=1e-6),metrics = ['accuracy'])
model.summary()

ResNet架構

12層

#12層 ResNet
tag = 'ResNet0910_fashion_mnist_12layerResNet_{}'.format(0)
h5_weight_path = os.path.join(WEIGHT_DIR, './' + tag + '.h5')

input_shape=(28,28,1)

input = Input(input_shape, name='input')

layer=Conv2D(32, kernel_size=(2, 2), activation='relu', padding='same')(input)
layer=BatchNormalization()(layer)
Resnet=layer
for i in range(2):
  layer=Conv2D(32, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=add([layer,Resnet])
layer=MaxPooling2D(pool_size=(2, 2))(layer)

Resnet=Conv2D(64, kernel_size=(1, 1), activation='relu',padding='same')(layer)
for i in range(3):
  layer=Conv2D(64, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=add([layer,Resnet])
layer=MaxPooling2D(pool_size=(2, 2))(layer)

Resnet=Conv2D(128, kernel_size=(1, 1), activation='relu',padding='same')(layer)
for i in range(3):
  layer=Conv2D(128, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=add([layer,Resnet])
layer=MaxPooling2D(pool_size=(2, 2))(layer)

Resnet=Conv2D(256, kernel_size=(1, 1), activation='relu',padding='same')(layer)
for i in range(3):
  layer=Conv2D(256, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=add([layer,Resnet])
layer=MaxPooling2D(pool_size=(2, 2))(layer)

layer = Dropout(0.5)(layer)
layer = Flatten(name='flatten')(layer)
output = Dense(nb_classes, name="Dense_10nb", activation='softmax')(layer)

model = Model(inputs=[input], outputs=[output])

model.compile(loss='sparse_categorical_crossentropy',optimizer=keras.optimizers.Adam(lr=0.0001,decay=1e-6),metrics = ['accuracy'])
model.summary()

20層

#20層 ResNet
tag = 'ResNet0910_fashion_mnist_20layer_ResNet_{}'.format(0)
h5_weight_path = os.path.join(WEIGHT_DIR, './' + tag + '.h5')

input_shape=(28,28,1)

input = Input(input_shape, name='input')

layer=Conv2D(32, kernel_size=(2, 2), activation='relu', padding='same')(input)
layer=BatchNormalization()(layer)
Resnet=layer
for i in range(4):
  layer=Conv2D(32, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=add([layer,Resnet])
layer=MaxPooling2D(pool_size=(2, 2))(layer)

Resnet=Conv2D(64, kernel_size=(1, 1), activation='relu',padding='same')(layer)
for i in range(5):
  layer=Conv2D(64, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=add([layer,Resnet])
layer=MaxPooling2D(pool_size=(2, 2))(layer)

Resnet=Conv2D(128, kernel_size=(1, 1), activation='relu',padding='same')(layer)
for i in range(5):
  layer=Conv2D(128, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=add([layer,Resnet])
layer=MaxPooling2D(pool_size=(2, 2))(layer)

Resnet=Conv2D(256, kernel_size=(1, 1), activation='relu',padding='same')(layer)
for i in range(5):
  layer=Conv2D(256, kernel_size=(2, 2), activation='relu',padding='same')(layer)
  layer=BatchNormalization()(layer)
layer=add([layer,Resnet])
layer=MaxPooling2D(pool_size=(2, 2))(layer)

layer = Dropout(0.5)(layer)
layer = Flatten(name='flatten')(layer)
output = Dense(nb_classes, name="Dense_10nb", activation='softmax')(layer)

model = Model(inputs=[input], outputs=[output])

model.compile(loss='sparse_categorical_crossentropy',optimizer=keras.optimizers.Adam(lr=0.0001,decay=1e-6),metrics = ['accuracy'])
model.summary()

沒加入ResNet正確率

在我們這種設計下，可以發現越多層數，我們的正確率越低。
那麼我們加入ResNet試試看正確率是否能提高。

ResNet架構正確率

可以發現使用ResNet後，正確率都有提升了!

結論

在沒加入ResNet時，12層正確率最好，而50層正確率最低，因為太過深層了!
使用ResNet後，每一種層數的正確率都有提升，尤其是50層的，從完全無法看，變得人模人樣了(誤。

參考資料

Deep Residual Learning for Image Recognition