深度学习笔记——分类模型（四）Inceptionv3

# -*- coding: utf-8 -*-
import datetime
from keras.models import Model
from keras.layers import *
from keras.preprocessing.image import ImageDataGenerator

def conv2d_bn(x, filters, num_row, num_col, padding='same', strides=(1, 1), name=None):
    if name is not None:
        bn_name = name + '_bn'
        conv_name = name + '_conv'
    else:
        bn_name = None
        conv_name = None
    bn_axis = 3

    x = Conv2D(filters, (num_row, num_col), strides=strides, padding=padding, use_bias=False, name=conv_name)(x)
    x = BatchNormalization(axis=bn_axis, scale=False, name=bn_name)(x)
    x = Activation('relu', name=name)(x)
    return x

def InceptionV3(input_shape=(299, 299, 3), classes=1000):
    img_input = Input(shape=input_shape)

    channel_axis = 3

    x = conv2d_bn(img_input, 32, 3, 3, strides=(2, 2), padding='valid')
    x = conv2d_bn(x, 32, 3, 3, padding='valid')
    x = conv2d_bn(x, 64, 3, 3)
    x = MaxPooling2D((3, 3), strides=(2, 2))(x)

    x = conv2d_bn(x, 80, 1, 1, padding='valid')
    x = conv2d_bn(x, 192, 3, 3, padding='valid')
    x = MaxPooling2D((3, 3), strides=(2, 2))(x)

    # mixed 0, 1, 2: 35 x 35 x 256
    branch1x1 = conv2d_bn(x, 64, 1, 1)

    branch5x5 = conv2d_bn(x, 48, 1, 1)
    branch5x5 = conv2d_bn(branch5x5, 64, 5, 5)

    branch3x3dbl = conv2d_bn(x, 64, 1, 1)
    branch3x3dbl = conv2d_bn(branch3x3dbl, 96, 3, 3)
    branch3x3dbl = conv2d_bn(branch3x3dbl, 96, 3, 3)

    branch_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')(x)
    branch_pool = conv2d_bn(branch_pool, 32, 1, 1)
    x = concatenate([branch1x1, branch5x5, branch3x3dbl, branch_pool], axis=channel_axis, name='mixed0')

    # mixed 1: 35 x 35 x 256
    branch1x1 = conv2d_bn(x, 64, 1, 1)

    branch5x5 = conv2d_bn(x, 48, 1, 1)
    branch5x5 = conv2d_bn(branch5x5, 64, 5, 5)

    branch3x3dbl = conv2d_bn(x, 64, 1, 1)
    branch3x3dbl = conv2d_bn(branch3x3dbl, 96, 3, 3)
    branch3x3dbl = conv2d_bn(branch3x3dbl, 96, 3, 3)

    branch_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')(x)
    branch_pool = conv2d_bn(branch_pool, 64, 1, 1)
    x = concatenate([branch1x1, branch5x5, branch3x3dbl, branch_pool], axis=channel_axis, name='mixed1')

    # mixed 2: 35 x 35 x 256
    branch1x1 = conv2d_bn(x, 64, 1, 1)

    branch5x5 = conv2d_bn(x, 48, 1, 1)
    branch5x5 = conv2d_bn(branch5x5, 64, 5, 5)

    branch3x3dbl = conv2d_bn(x, 64, 1, 1)
    branch3x3dbl = conv2d_bn(branch3x3dbl, 96, 3, 3)
    branch3x3dbl = conv2d_bn(branch3x3dbl, 96, 3, 3)

    branch_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')(x)
    branch_pool = conv2d_bn(branch_pool, 64, 1, 1)
    x = concatenate([branch1x1, branch5x5, branch3x3dbl, branch_pool], axis=channel_axis, name='mixed2')

    # mixed 3: 17 x 17 x 768
    branch3x3 = conv2d_bn(x, 384, 3, 3, strides=(2, 2), padding='valid')

    branch3x3dbl = conv2d_bn(x, 64, 1, 1)
    branch3x3dbl = conv2d_bn(branch3x3dbl, 96, 3, 3)
    branch3x3dbl = conv2d_bn(branch3x3dbl, 96, 3, 3, strides=(2, 2), padding='valid')

    branch_pool = MaxPooling2D((3, 3), strides=(2, 2))(x)
    x = concatenate([branch3x3, branch3x3dbl, branch_pool], axis=channel_axis, name='mixed3')

    # mixed 4: 17 x 17 x 768
    branch1x1 = conv2d_bn(x, 192, 1, 1)

    branch7x7 = conv2d_bn(x, 128, 1, 1)
    branch7x7 = conv2d_bn(branch7x7, 128, 1, 7)
    branch7x7 = conv2d_bn(branch7x7, 192, 7, 1)

    branch7x7dbl = conv2d_bn(x, 128, 1, 1)
    branch7x7dbl = conv2d_bn(branch7x7dbl, 128, 7, 1)
    branch7x7dbl = conv2d_bn(branch7x7dbl, 128, 1, 7)
    branch7x7dbl = conv2d_bn(branch7x7dbl, 128, 7, 1)
    branch7x7dbl = conv2d_bn(branch7x7dbl, 192, 1, 7)

    branch_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')(x)
    branch_pool = conv2d_bn(branch_pool, 192, 1, 1)
    x = concatenate([branch1x1, branch7x7, branch7x7dbl, branch_pool], axis=channel_axis, name='mixed4')

    # mixed 5, 6: 17 x 17 x 768
    for i in range(2):
        branch1x1 = conv2d_bn(x, 192, 1, 1)

        branch7x7 = conv2d_bn(x, 160, 1, 1)
        branch7x7 = conv2d_bn(branch7x7, 160, 1, 7)
        branch7x7 = conv2d_bn(branch7x7, 192, 7, 1)

        branch7x7dbl = conv2d_bn(x, 160, 1, 1)
        branch7x7dbl = conv2d_bn(branch7x7dbl, 160, 7, 1)
        branch7x7dbl = conv2d_bn(branch7x7dbl, 160, 1, 7)
        branch7x7dbl = conv2d_bn(branch7x7dbl, 160, 7, 1)
        branch7x7dbl = conv2d_bn(branch7x7dbl, 192, 1, 7)

        branch_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')(x)
        branch_pool = conv2d_bn(branch_pool, 192, 1, 1)
        x = concatenate([branch1x1, branch7x7, branch7x7dbl, branch_pool], axis=channel_axis, name='mixed' + str(5 + i))

    # mixed 7: 17 x 17 x 768
    branch1x1 = conv2d_bn(x, 192, 1, 1)

    branch7x7 = conv2d_bn(x, 192, 1, 1)
    branch7x7 = conv2d_bn(branch7x7, 192, 1, 7)
    branch7x7 = conv2d_bn(branch7x7, 192, 7, 1)

    branch7x7dbl = conv2d_bn(x, 192, 1, 1)
    branch7x7dbl = conv2d_bn(branch7x7dbl, 192, 7, 1)
    branch7x7dbl = conv2d_bn(branch7x7dbl, 192, 1, 7)
    branch7x7dbl = conv2d_bn(branch7x7dbl, 192, 7, 1)
    branch7x7dbl = conv2d_bn(branch7x7dbl, 192, 1, 7)

    branch_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')(x)
    branch_pool = conv2d_bn(branch_pool, 192, 1, 1)
    x = concatenate([branch1x1, branch7x7, branch7x7dbl, branch_pool], axis=channel_axis, name='mixed7')

    # mixed 8: 8 x 8 x 1280
    branch3x3 = conv2d_bn(x, 192, 1, 1)
    branch3x3 = conv2d_bn(branch3x3, 320, 3, 3,
                          strides=(2, 2), padding='valid')

    branch7x7x3 = conv2d_bn(x, 192, 1, 1)
    branch7x7x3 = conv2d_bn(branch7x7x3, 192, 1, 7)
    branch7x7x3 = conv2d_bn(branch7x7x3, 192, 7, 1)
    branch7x7x3 = conv2d_bn(branch7x7x3, 192, 3, 3, strides=(2, 2), padding='valid')

    branch_pool = MaxPooling2D((3, 3), strides=(2, 2))(x)
    x = concatenate([branch3x3, branch7x7x3, branch_pool], axis=channel_axis, name='mixed8')

    # mixed 9: 8 x 8 x 2048
    for i in range(2):
        branch1x1 = conv2d_bn(x, 320, 1, 1)

        branch3x3 = conv2d_bn(x, 384, 1, 1)
        branch3x3_1 = conv2d_bn(branch3x3, 384, 1, 3)
        branch3x3_2 = conv2d_bn(branch3x3, 384, 3, 1)
        branch3x3 = concatenate([branch3x3_1, branch3x3_2], axis=channel_axis, name='mixed9_' + str(i))

        branch3x3dbl = conv2d_bn(x, 448, 1, 1)
        branch3x3dbl = conv2d_bn(branch3x3dbl, 384, 3, 3)
        branch3x3dbl_1 = conv2d_bn(branch3x3dbl, 384, 1, 3)
        branch3x3dbl_2 = conv2d_bn(branch3x3dbl, 384, 3, 1)
        branch3x3dbl = concatenate([branch3x3dbl_1, branch3x3dbl_2], axis=channel_axis)

        branch_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')(x)
        branch_pool = conv2d_bn(branch_pool, 192, 1, 1)
        x = concatenate([branch1x1, branch3x3, branch3x3dbl, branch_pool], axis=channel_axis, name='mixed' + str(9 + i))

    x = GlobalAveragePooling2D(name='avg_pool')(x)
    if classes == 2:
        x = Dense(1, activation='sigmoid', name='predictions')(x)
    else:
        x = Dense(classes, activation='softmax', name='predictions')(x)
    # Create model.
    model = Model(img_input, x, name='inception_v3')

    return model

def main():
    width = 299
    height = 299
    batch_size = 2

    generator = ImageDataGenerator(horizontal_flip=True,
                                   vertical_flip=True,
                                   validation_split=0.2)

    train_generator = generator.flow_from_directory(directory="datasets/train",
                                                    target_size=(width, height),
                                                    batch_size=batch_size,
                                                    class_mode="binary",
                                                    subset="training")

    val_generator = generator.flow_from_directory(directory="datasets/train",
                                                  target_size=(width, height),
                                                  batch_size=batch_size,
                                                  class_mode="binary",
                                                  subset="validation")

    model = InceptionV3(classes=2)
    model.summary()
    # Compile model
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])

    model.fit_generator(train_generator, validation_data=val_generator, epochs=10, verbose=1)

if __name__ == '__main__':
    tic = datetime.datetime.now()
    main()
    toc = datetime.datetime.now()
    print("\nThis model took ", (toc - tic))