InceptionV3的keras实现

最新推荐文章于 2023-05-22 17:12:42 发布
最新推荐文章于 2023-05-22 17:12:42 发布
阅读量804 收藏 5
点赞数 1
分类专栏： ML&DL 文章标签： InceptionV3 Keras
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/liugan528/article/details/123759406
版权

InceptionV3.py

from tensorflow import keras
from tensorflow.keras import layers

def InceptionBlockA(input, pooling_channel, idx):
    x1 = layers.Conv2D(64, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_a'+str(idx))(input)
    x1 = layers.BatchNormalization(scale=False, name='inception_a'+str(idx)+'_1x1_bn')(x1)
    x1 = layers.Activation('relu', name='inception_a'+str(idx)+'_1x1_relu')(x1)

    x2 = layers.Conv2D(48, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_a'+str(idx)+'_5x5_reduce')(input)
    x2 = layers.BatchNormalization(scale=False, name='inception_a'+str(idx)+'_5x5_reduce_bn')(x2)
    x2 = layers.Activation('relu', name='inception_a'+str(idx)+'_5x5_reduce_relu')(x2)
    x2 = layers.Conv2D(64, 5, strides=(1, 1), padding='same', use_bias=False, name='inception_a'+str(idx)+'_5x5')(x2)
    x2 = layers.BatchNormalization(scale=False, name='inception_a'+str(idx)+'_5x5_bn')(x2)
    x2 = layers.Activation('relu', name='inception_a'+str(idx)+'_5x5_relu')(x2)

    x3 = layers.Conv2D(64, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_a'+str(idx)+'_3x3_reduce')(input)
    x3 = layers.BatchNormalization(scale=False, name='inception_a'+str(idx)+'_3x3_reduce_bn')(x3)
    x3 = layers.Activation('relu', name='inception_a'+str(idx)+'_3x3_reduce_relu')(x3)
    x3 = layers.Conv2D(96, 3, strides=(1, 1), padding='same', use_bias=False, name='inception_a'+str(idx)+'_3x3_1')(x3)
    x3 = layers.BatchNormalization(scale=False, name='inception_a'+str(idx)+'_3x3_1_bn')(x3)
    x3 = layers.Activation('relu', name='inception_a'+str(idx)+'_3x3_1_relu')(x3)
    x3 = layers.Conv2D(96, 3, strides=(1, 1), padding='same', use_bias=False, name='inception_a'+str(idx)+'_3x3_2')(x3)
    x3 = layers.BatchNormalization(scale=False, name='inception_a'+str(idx)+'_3x3_2_bn')(x3)
    x3 = layers.Activation('relu', name='inception_a'+str(idx)+'_3x3_2_relu')(x3)

    x4 = layers.AveragePooling2D((3, 3), strides=(1, 1), padding='same', name='inception_a'+str(idx)+'_pool')(input)
    x4 = layers.Conv2D(pooling_channel, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_a'+str(idx)+'_pool_proj')(x4)
    x4 = layers.BatchNormalization(scale=False, name='inception_a'+str(idx)+'_pool_proj_bn')(x4)
    x4 = layers.Activation('relu', name='inception_a'+str(idx)+'_pool_proj_relu')(x4)

    return layers.concatenate([x1, x2, x3, x4], name='inception_a'+str(idx)+'_output')


def ReductionA(input):
    x1 = layers.Conv2D(384, 3, strides=(2, 2), padding='valid', use_bias=False, name='reduction_a_3x3')(input)
    x1 = layers.BatchNormalization(scale=False, name='reduction_a_3x3_bn')(x1)
    x1 = layers.Activation('relu', name='reduction_a_3x3_relu')(x1)

    x2 = layers.Conv2D(64, 1, strides=(1, 1), padding='valid', use_bias=False, name='reduction_a_3x3_2_reduce')(input)
    x2 = layers.BatchNormalization(scale=False, name='reduction_a_3x3_2_reduce_bn')(x2)
    x2 = layers.Activation('relu', name='reduction_a_3x3_2_reduce_relu')(x2)
    x2 = layers.Conv2D(96, 3, strides=(1, 1), padding='same', use_bias=False, name='reduction_a_3x3_2')(x2)
    x2 = layers.BatchNormalization(scale=False, name='reduction_a_3x3_2_bn')(x2)
    x2 = layers.Activation('relu', name='reduction_a_3x3_2_relu')(x2)
    x2 = layers.Conv2D(96, 3, strides=(2, 2), padding='valid', use_bias=False, name='reduction_a_3x3_3')(x2)
    x2 = layers.BatchNormalization(scale=False, name='reduction_a_3x3_3_bn')(x2)
    x2 = layers.Activation('relu', name='reduction_a_3x3_3_relu')(x2)

    x3 = layers.MaxPooling2D((3, 3), strides=(2, 2), padding='valid', name='reduction_a_pool')(input)

    return layers.concatenate([x1, x2, x3], name='reduction_a_concat')


def InceptionBlockB(input, x2_channel=[128, 128, 192], x3_channel=[128, 128, 128, 128, 192], idx=1):
    x1 = layers.Conv2D(192, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'1_1x1_2')(input)
    x1 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'1_1x1_2_bn')(x1)
    x1 = layers.Activation('relu', name='inception_b'+str(idx)+'1_1x1_2_relu')(x1)

    x2 = layers.Conv2D(x2_channel[0], 1, strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'_1x7_reduce')(input)
    x2 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'_1x7_reduce_bn')(x2)
    x2 = layers.Activation('relu', name='inception_b'+str(idx)+'_1x7_reduce_relu')(x2)
    x2 = layers.Conv2D(x2_channel[1], (7, 1), strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'_1x7')(x2)
    x2 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'_1x7_bn')(x2)
    x2 = layers.Activation('relu', name='inception_b'+str(idx)+'_1x7_relu')(x2)
    x2 = layers.Conv2D(x2_channel[2], (1, 7), strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'_7x1')(x2)
    x2 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'_7x1_bn')(x2)
    x2 = layers.Activation('relu', name='inception_b'+str(idx)+'_7x1_relu')(x2)

    x3 = layers.Conv2D(x3_channel[0], 1, strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'_7x1_reduce')(input)
    x3 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'_7x1_reduce_bn')(x3)
    x3 = layers.Activation('relu', name='inception_b'+str(idx)+'_7x1_reduce_relu')(x3)
    x3 = layers.Conv2D(x3_channel[1], (1, 7), strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'_7x1_2')(x3)
    x3 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'_7x1_2_bn')(x3)
    x3 = layers.Activation('relu', name='inception_b'+str(idx)+'_7x1_2_relu')(x3)
    x3 = layers.Conv2D(x3_channel[2], (7, 1), strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'_1x7_2')(x3)
    x3 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'_1x7_2_bn')(x3)
    x3 = layers.Activation('relu', name='inception_b'+str(idx)+'_1x7_2_relu')(x3)
    x3 = layers.Conv2D(x3_channel[3], (1, 7), strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'_7x1_3')(x3)
    x3 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'_7x1_3_bn')(x3)
    x3 = layers.Activation('relu', name='inception_b'+str(idx)+'_7x1_3_relu')(x3)
    x3 = layers.Conv2D(x3_channel[4], (7, 1), strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'_1x7_3')(x3)
    x3 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'_1x7_3_bn')(x3)
    x3 = layers.Activation('relu', name='inception_b'+str(idx)+'_1x7_3_relu')(x3)

    x4 = layers.AveragePooling2D((3, 3), strides=(1, 1), padding='same', name='inception_b'+str(idx)+'_pool_ave')(input)
    x4 = layers.Conv2D(192, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_b'+str(idx)+'_1x1')(x4)
    x4 = layers.BatchNormalization(scale=False, name='inception_b'+str(idx)+'_1x1_bn')(x4)
    x4 = layers.Activation('relu', name='inception_b'+str(idx)+'_1x1_relu')(x4)

    return layers.concatenate([x1, x2, x3, x4], name='inception_b'+str(idx)+'_concat')


def ReductionB(input):
    x1 = layers.Conv2D(192, 1, strides=(1, 1), padding='valid', use_bias=False, name='reduction_b_3x3_reduce')(input)
    x1 = layers.BatchNormalization(scale=False, name='reduction_b_3x3_reduce_bn')(x1)
    x1 = layers.Activation('relu', name='reduction_b_3x3_reduce_relu')(x1)
    x1 = layers.Conv2D(320, 3, strides=(2, 2), padding='valid', use_bias=False, name='reduction_b_3x3')(x1)
    x1 = layers.BatchNormalization(scale=False, name='reduction_b_3x3_bn')(x1)
    x1 = layers.Activation('relu', name='reduction_b_3x3_relu')(x1)

    x2 = layers.Conv2D(192, 1, strides=(1, 1), padding='valid', use_bias=False, name='reduction_b_1x7_reduce')(input)
    x2 = layers.BatchNormalization(scale=False, name='reduction_b_1x7_reduce_bn')(x2)
    x2 = layers.Activation('relu', name='reduction_b_1x7_reduce_relu')(x2)
    x2 = layers.Conv2D(192, (7, 1), strides=(1, 1), padding='same', use_bias=False, name='reduction_b_1x7')(x2)
    x2 = layers.BatchNormalization(scale=False, name='reduction_b_1x7_bn')(x2)
    x2 = layers.Activation('relu', name='reduction_b_1x7_relu')(x2)
    x2 = layers.Conv2D(192, (1, 7), strides=(1, 1), padding='same', use_bias=False, name='reduction_b_7x1')(x2)
    x2 = layers.BatchNormalization(scale=False, name='reduction_b_7x1_bn')(x2)
    x2 = layers.Activation('relu', name='reduction_b_7x1_relu')(x2)
    x2 = layers.Conv2D(192, 3, strides=(2, 2), padding='valid', use_bias=False, name='reduction_b_3x3_2')(x2)
    x2 = layers.BatchNormalization(scale=False, name='reduction_b_3x3_2_bn')(x2)
    x2 = layers.Activation('relu', name='reduction_b_3x3_2_relu')(x2)

    x3 = layers.MaxPooling2D((3, 3), strides=(2, 2), padding='valid', name='reduction_b_pool')(input)

    return layers.concatenate([x1, x2, x3], name='reduction_b_concat')


def InceptionBlockC(input, idx):
    x1 = layers.Conv2D(320, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_c'+str(idx)+'_1x1_2')(input)
    x1 = layers.BatchNormalization(scale=False, name='inception_c'+str(idx)+'_1x1_2_bn')(x1)
    x1 = layers.Activation('relu', name='inception_c'+str(idx)+'_1x1_2_relu')(x1)

    tmp1 = layers.Conv2D(384, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_c'+str(idx)+'_1x3_reduce')(input)
    tmp1 = layers.BatchNormalization(scale=False, name='inception_c'+str(idx)+'_1x3_reduce_bn')(tmp1)
    tmp1 = layers.Activation('relu', name='inception_c'+str(idx)+'_1x3_reduce_relu')(tmp1)

    x2 = layers.Conv2D(384, (3, 1), strides=(1, 1), padding='same', use_bias=False, name='inception_c'+str(idx)+'_1x3')(tmp1)
    x2 = layers.BatchNormalization(scale=False, name='inception_c'+str(idx)+'_1x3_bn')(x2)
    x2 = layers.Activation('relu', name='inception_c'+str(idx)+'_1x3_relu')(x2)

    x3 = layers.Conv2D(384, (1, 3), strides=(1, 1), padding='same', use_bias=False, name='inception_c'+str(idx)+'_3x1')(tmp1)
    x3 = layers.BatchNormalization(scale=False, name='inception_c'+str(idx)+'_3x1_bn')(x3)
    x3 = layers.Activation('relu', name='inception_c'+str(idx)+'_3x1_relu')(x3)

    tmp2 = layers.Conv2D(448, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_c'+str(idx)+'_3x3_reduce')(input)
    tmp2 = layers.BatchNormalization(scale=False, name='inception_c'+str(idx)+'_3x3_reduce_bn')(tmp2)
    tmp2 = layers.Activation('relu', name='inception_c'+str(idx)+'_3x3_reduce_relu')(tmp2)
    tmp2 = layers.Conv2D(384, 3, strides=(1, 1), padding='same', use_bias=False, name='inception_c'+str(idx)+'_3x3')(tmp2)
    tmp2 = layers.BatchNormalization(scale=False, name='inception_c'+str(idx)+'_3x3_bn')(tmp2)
    tmp2 = layers.Activation('relu', name='inception_c'+str(idx)+'_3x3_relu')(tmp2)

    x4 = layers.Conv2D(384, (3, 1), strides=(1, 1), padding='same', use_bias=False, name='inception_c'+str(idx)+'_1x3_2')(tmp2)
    x4 = layers.BatchNormalization(scale=False, name='inception_c'+str(idx)+'_1x3_2_bn')(x4)
    x4 = layers.Activation('relu', name='inception_c'+str(idx)+'_1x3_2_relu')(x4)

    x5 = layers.Conv2D(384, (1, 3), strides=(1, 1), padding='same', use_bias=False, name='inception_c'+str(idx)+'_3x1_2')(tmp2)
    x5 = layers.BatchNormalization(scale=False, name='inception_c'+str(idx)+'_3x1_2_bn')(x5)
    x5 = layers.Activation('relu', name='inception_c'+str(idx)+'_3x1_2_relu')(x5)

    x6 = layers.AveragePooling2D((3, 3), strides=(1, 1), padding='same', name='inception_c'+str(idx)+'_pool')(input)
    x6 = layers.Conv2D(192, 1, strides=(1, 1), padding='same', use_bias=False, name='inception_c'+str(idx)+'_1x1')(x6)
    x6 = layers.BatchNormalization(scale=False, name='inception_c'+str(idx)+'_1x1_bn')(x6)
    x6 = layers.Activation('relu', name='inception_c'+str(idx)+'_1x1_relu')(x6)

    return layers.concatenate([x1, x2, x3, x4, x5, x6], name='inception_c'+str(idx)+'_concat')


def InceptionV3(input_shape, output_class):
    input = keras.Input(shape=input_shape, name='input')
    x = layers.Conv2D(32, 3, strides=(2, 2), padding='valid', use_bias=False, name='conv1_3x3_s2')(input)
    x = layers.BatchNormalization(scale=False, name='conv1_3x3_s2_bn')(x)
    x = layers.Activation('relu', name='conv1_3x3_relu')(x)

    x = layers.Conv2D(32, 3, strides=(1, 1), padding='valid', use_bias=False, name='conv2_3x3_s1')(x)
    x = layers.BatchNormalization(scale=False, name='conv2_3x3_s1_bn')(x)
    x = layers.Activation('relu', name='conv2_3x3_relu')(x)

    x = layers.Conv2D(64, 3, strides=(1, 1), padding='same', use_bias=False, name='conv3_3x3_s1')(x)
    x = layers.BatchNormalization(scale=False, name='conv3_3x3_s1_bn')(x)
    x = layers.Activation('relu', name='conv3_3x3_relu')(x)

    x = layers.MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='valid', name='pool1_3x3_s2')(x)

    x = layers.Conv2D(80, 1, strides=(1, 1), padding='valid', use_bias=False, name='conv4_3x3_reduce')(x)
    x = layers.BatchNormalization(scale=False, name='conv4_3x3_reduce_bn')(x)
    x = layers.Activation('relu', name='conv4_relu_3x3_reduce')(x)

    x = layers.Conv2D(192, 3, strides=(1, 1), padding='valid', use_bias=False, name='conv4_3x3')(x)
    x = layers.BatchNormalization(scale=False, name='conv4_3x3_bn')(x)
    x = layers.Activation('relu', name='conv4_relu_3x3')(x)

    x = layers.MaxPooling2D((3, 3), strides=(2, 2), padding='valid', name='pool2_3x3_s2')(x)

    x = InceptionBlockA(x, 32, 1)
    x = InceptionBlockA(x, 64, 2)
    x = InceptionBlockA(x, 64, 3)

    x = ReductionA(x)

    x = InceptionBlockB(x, x2_channel=[128, 128, 192], x3_channel=[128, 128, 128, 128, 192], idx=1)
    x = InceptionBlockB(x, x2_channel=[160, 160, 192], x3_channel=[160, 160, 160, 160, 192], idx=2)
    x = InceptionBlockB(x, x2_channel=[160, 160, 192], x3_channel=[160, 160, 160, 160, 192], idx=3)
    x = InceptionBlockB(x, x2_channel=[192, 192, 192], x3_channel=[192, 192, 192, 192, 192], idx=4)

    x = ReductionB(x)

    x = InceptionBlockC(x, 1)
    x = InceptionBlockC(x, 2)

    # x = layers.AveragePooling2D((8, 8), name='pool_8x8_s1')(x)
    # x = layers.Flatten(name='flatten')(x)
    x = layers.GlobalAveragePooling2D(name='pool_8x8_s1')(x)
    x = layers.Dense(output_class, activation='softmax', name='dense')(x)

    model = keras.Model(input, x, name='inceptionv3')

    model.summary()
    return model

if __name__ == '__main__':
    model = InceptionV3((299, 299, 3), 1000)
    print('Done.')

train_inceptionv3.py

import numpy as np
import cv2
from InceptionV3 import InceptionV3
from tensorflow.keras import backend as K   # K.set_image_dim_ordering('tf')
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping, LearningRateScheduler, ReduceLROnPlateau
from tensorflow.keras import optimizers
import matplotlib.pyplot as plt

NUM_CLASSES = 2
INPUT_IMG_SIZE = 299

def show_train_history(train_history, train, validation):
    plt.plot(train_history.history[train])
    plt.plot(train_history.history[validation])
    plt.title('Train history')
    plt.ylabel(train)
    plt.xlabel('Epoch')
    plt.legend(['train', 'validation'], loc='upper left')
    plt.show()

def generate_arrays_from_file(root_path, lines, num_classes, batch_size):
    num = len(lines)   # 数据总个数
    i = 0
    while 1:
        x_images_normalize = []
        y_labels_onehot = []

        # 获取一个batch_size大小的数据
        for b in range(batch_size):
            if i == 0:
                np.random.shuffle(lines)
            img_path = root_path + "\\image\\train\\" + lines[i].split(";")[0]  # cat dog datasets
            # img_path = root_path + lines[i].split(" ")[0]                     # FlowerPhotos datasets
            img = cv2.imread(img_path)
            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
            img = cv2.resize(img, (INPUT_IMG_SIZE, INPUT_IMG_SIZE))
            img = img / 255
            x_images_normalize.append(img)
            label = to_categorical(lines[i].split(";")[1], num_classes=num_classes)    # cat dog datasets
            # label = to_categorical(lines[i].split(" ")[1], num_classes=num_classes)  # FlowerPhotos datasets
            y_labels_onehot.append(label)
            i = (i+1) % num  # 遍历一个epoch重新打乱数据
        x_images_normalize = np.array(x_images_normalize)
        #x_images_normalize = x_images_normalize.reshape(-1, 224, 224, 3)
        y_labels_onehot = np.array(y_labels_onehot)
        yield (x_images_normalize, y_labels_onehot)

if __name__ == "__main__":
    anchors = np.zeros((9, 4))

    lines = []

    ### cat dog dataset
    root_path = r"D:\03.Data\01.CatDog"
    with open(root_path + "\\dataset.txt") as f:
       lines = f.readlines()

    ### FlowerPhotos dataset
    # root_path = r"D:\03.Data\02.FlowerPhotos"
    # with open(root_path + "\\train.txt") as f:
    #   lines = f.readlines()

    np.random.shuffle(lines)  # 随机打乱数据
    val_ratio = 0.1
    val_num = int(len(lines) * val_ratio)
    val_lines = lines[0:val_num]
    train_lines = lines[val_num:]

    model = InceptionV3((INPUT_IMG_SIZE, INPUT_IMG_SIZE, 3), NUM_CLASSES)

    # 保存的方式，3个epoch保存一次
    # https://keras-zh.readthedocs.io/callbacks/#modelcheckpoint
    log_path = r"D:\02.Work\00.LearnML\003.Net\InceptionV3\log\\"
    checkpoint = ModelCheckpoint(log_path + "weights.epoch{epoch:03d}-loss{loss:.3f}-val_loss{val_loss:.3f}.hdf5",
                                 monitor='accuracy',  # loss或accuracy或val_loss或val_accuracy或lr
                                 verbose=0,
                                 save_best_only=False,  #  True
                                 save_weights_only=False,
                                 period=1)

    # 学习率下降的方式，acc三次不下降就下降学习率继续训练
    # https://keras-zh.readthedocs.io/callbacks/#reducelronplateau
    reduce_lr = ReduceLROnPlateau(monitor='accuracy', factor=0.5, patience=3) # monitor=loss、accuracy、val_loss、val_accuracy、lr

    # 当val_loss一直不下降的时候意味着模型基本训练完毕，可以停止
    # https://keras-zh.readthedocs.io/callbacks/#earlystopping
    early_stopping = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=10, verbose=0)

    # 交叉熵损失
    # opti = optimizers.SGD(lr=0.001, decay=1e-6, momentum=0.9, nesterov=True)
    opti = optimizers.Adam(lr=1e-3)
    model.compile(optimizer=opti, loss='categorical_crossentropy', metrics=['accuracy'])

    batch_size = 128//8
    print('Train on {} samples, val on {} samples, with batch size {}.'.format(len(train_lines), len(val_lines), batch_size))

    # 开始训练
    train_history = model.fit_generator(generate_arrays_from_file(root_path, train_lines, NUM_CLASSES, batch_size),
                                        steps_per_epoch=max(len(train_lines)//batch_size, 1),
                                        epochs=10,
                                        verbose=1,
                                        callbacks=[checkpoint, reduce_lr, early_stopping],
                                        validation_data=generate_arrays_from_file(root_path, val_lines, NUM_CLASSES, batch_size),
                                        validation_steps=max(len(val_lines)//batch_size, 1),
                                        validation_freq=1,
                                        initial_epoch=0)

    # 保存模型
    model.save_weights(log_path+'inceptionv3.h5')

    show_train_history(train_history, 'accuracy', 'val_accuracy')
    show_train_history(train_history, 'loss', 'val_loss')
    print('Done.')

    # 评估模型准确率
    # scores = model.evaluate(x_img_test_normalize, y_label_test_onehot, verbose=0)
    # print(len(scores))

    # 进行预测
    # prediction = model.predict_classes(x_test_image_normalize)
    # prediction[:10]   # 查看预测结果的前10项数据

predict_inceptionv3.py

import matplotlib.pyplot as plt

from InceptionV3 import InceptionV3
import cv2
import numpy as np
from tensorflow.keras import backend as K   # K.set_image_dim_ordering('tf')
from tensorflow.keras.utils import to_categorical

INPUT_IMG_SIZE = 299
NUM_CLASSES = 2
label_dict = {0:'CAT', 1:'DOG'}

def show_predict_probability(y_gts, predictions, x_imgs, predict_probabilitys, idx):
    for i in range(len(label_dict)):
        print(label_dict[i]+', Probability:%1.9f'%(predict_probabilitys[idx][i]))
    print('label: ', label_dict[int(y_gts[idx])], ', predict: ', label_dict[predictions[idx]])
    plt.figure(figsize=(2, 2))
    plt.imshow(np.reshape(x_imgs[idx], (INPUT_IMG_SIZE, INPUT_IMG_SIZE, 3)))
    plt.show()

def plot_images_labels_prediction(images, labels, prediction, idx, num):
    fig = plt.gcf()
    fig.set_size_inches(12, 14)
    if num>25: num=25
    for i in range(0, num):
        ax = plt.subplot(2, 5, 1+i)
        ax.imshow(images[idx], cmap='binary')
        title = 'labels='+str(labels[idx])
        if len(prediction) > 0:
            title += "prediction="+str(prediction[idx])
        ax.set_title(title, fontsize=10)
        idx += 1
    plt.show()

if __name__ == '__main__':
    log_path = r"D:\02.Work\00.LearnML\003.Net\InceptionV3\log\\"
    model = InceptionV3((299, 299, 3), NUM_CLASSES)
    #model.load_weights(log_path+"inceptionv3.h5")
    model.load_weights(log_path+"weights.epoch008-loss0.125-val_loss0.164.hdf5")

    ### cat dog dataset
    lines = []
    root_path = r"D:\03.Data\01.CatDog"
    with open(root_path + "\\test.txt") as f:
       lines = f.readlines()

    x_images_normalize = []
    y_labels_onehot = []
    y_labels = []

    for i in range(len(lines)):
        img_path = lines[i].split(";")[0]
        img = cv2.imread(img_path)
        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        img = cv2.resize(img, (INPUT_IMG_SIZE, INPUT_IMG_SIZE))
        img = img / 255
        x_images_normalize.append(img)
        label = to_categorical(lines[i].split(";")[1], num_classes=NUM_CLASSES)
        y_labels_onehot.append(label)
        y_labels.append(lines[i].split(";")[1])
    x_images_normalize = np.array(x_images_normalize)
    # x_images_normalize = x_images_normalize.reshape(-1, INPUT_IMG_SIZE, INPUT_IMG_SIZE, 3)
    y_labels_onehot = np.array(y_labels_onehot)

    predict_probability = model.predict(x_images_normalize, verbose=1)
    predict = np.argmax(predict_probability, axis=1)

    plot_images_labels_prediction(x_images_normalize, y_labels, predict, 0, 10)
    show_predict_probability(y_labels, predict, x_images_normalize, predict_probability, 0)
    print('done')
GarryLau
关注
  • 1
    点赞
  • 踩
  • 5
    收藏
    觉得还不错? 一键收藏
  • 1
    评论
专栏目录
SE-Inception v3架构的模型搭建（keras代码实现）
Mark_Lewis的博客
09-28 3510
首先，先上SENet架构的原理图： 图是将SE模块嵌入到Inception结构的一个示例。方框旁边的维度信息代表该层的输出。这里我们使用global average pooling作为Squeeze操作。紧接着两个Fully Connected 层组成一个Bottleneck结构去建模通道间的相关性，并输出和输入特征同样数目的权重。我们首先将特征维度降低到输入的1/16，然后经过ReLu激活...
深度学习Keras对InceptionV3迁移学习精讲
10-15
本课程讲解内容是基于深度学习框架Keras，对InceptionV3模型进行迁移学习。涉及到迁移学习的必要性，迁移学习方法，迁移学习实战，最后用迁移学习结果去识别图片。
InceptionV1网络结构讲解（Tensorflow-2.6.0实现结构）
Keep_Trying_Go的博客
04-19 2417
文章目录1.结构表2.InceptionV1提升网络性能的方法3.InceptionV1模块（1）模块一（2）模块二 （改进模块）4.网络结构5.关于1x1的作用6.Multi-scale processing 1.结构表 2.InceptionV1提升网络性能的方法 传统的模块提升网络性能的方法是增加网络深度和宽度（卷积核的个数），但是会存在一些问题： 1.参数量太大，如果训练数据集有限，很容易产生过拟合。 2.网络模型越大，参数越多，则计算复杂度越大，对内存的占用和计算量的要求很高。 3.网络结构越
InceptionV3详细代码,带准确率和loss分析，以及ROC曲线（python）
热门推荐
qq_38410428的博客
01-08 1万+
from __future__ import absolute_import from __future__ import division from __future__ import print_function import time start_time = time.time() import numpy as np import matplotlib.pyplot as plt from keras.callbacks import Callback,ModelCheckpoint from k
InceptionNet10详细原理（含tensorflow版源码）
Code_and516的博客
03-28 853
InceptionNet10详细原理（含tensorflow版源码）
keras实现GoogLeNet-InceptionV1
01-06
GoogLeNet Inception v1 结构 及 pytorch、tensorflow、keras、paddle实现ImageNet识别 ...# @File : InceptionV1-keras.py import keras from keras.utils import Sequence from keras.layers import * from
Python-Keras实现Inceptionv4InceptionResnetv1和v2网络架构
08-11
Keras实现Inception-v4, Inception - Resnet-v1和v2网络架构
tensorflow实现GoogLeNet-InceptionV1
01-06
GoogLeNet Inception v1 结构 及 pytorch、tensorflow、keras、paddle实现ImageNet识别 环境 python3.6，tensorflow-gpu 1.12.0 代码 # -*- coding: utf-8 -*- # @Time : 2020/2/3 9:56 # @Author : Zhao HL # @...
keras使用迁移学习实现医学图像二分类
08-31
使用keras建立InceptionV3基本模型，不包括顶层，使用预训练权重，在基本模型的基础上自定义几层神经网络，得到最后的模型，对模型进行训练 优化模型，调整超参数，提高准确率 在测试集上对模型进行评估，使用精确率...
backbones:定制骨干的keras实现
03-19
resnet34，resnet50，resnet101，resnext，resnest，se-resne * t，随机深度，CSP-resnet，2.5D，SENet，SKNet MobileNet mobileNetV1，mobileNetV2，mobileNetV3适用于deeplab的GoogleNet InceptionV1-4，...
keras_inception_v4_finetune.py例子
01-28
官方例子，深度学习专用，机器学习专用，代码简单，一看就会(keras inception v4 finetune)
inception_v3_keras.zip
01-03
keras的inception_v3预训练模型，国内下载比较麻烦，好不容易下下来，分享一下，使用方法写了个简单的教程，压缩包中有三个文件： > inception_v3_weights_tf_dim_ordering_tf_kernels.h5 #预训练模型文件 > inceptionv3.py #示例代码 > cropped_panda.jpg #示例图片
keras_inception_v3_finetune.py例子
01-28
官方例子，深度学习专用，机器学习专用，代码简单，一看就会(keras inception v3 finetune)
keras-finetuning, 使用你自己的数据集训练基于InceptionV3的图像分类器.zip
09-17
keras-finetuning, 使用你自己的数据集训练基于InceptionV3的图像分类器 使用你自己的数据集训练InceptionV3-based图像分类器基于在的新集合中的微调( InceptionV3 )，在 https://keras.io/applications/中的例子依赖项最新的( 来自源的> =1.0
GoogLeNet之InceptionV3神经网络简介与代码实战
最新发布
修行之路
05-22 314
GoogLeNet之InceptionV3神经网络简介与代码实战
使用TensorFlow Keras手动搭建InceptionV3网络
清风专栏
07-11 1177
导入函数库 import tensorflow as tf import numpy as np import tensorflow_datasets as tfds # 这个是指Tensorflow Datasets import matplotlib.pyplot as plt 定义网络结构 一些参数设置 layers = tf.keras.layers models = tf.keras.models 定义InceptionV3模型用于caltech101物体分类 def conv2d_bn(x,
【keras实战】用Inceptionv3实现五种花的分类
m0_37935211的博客
10-10 7746
参考：keras系列︱迁移学习：利用InceptionV3进行fine-tuning及预测、完美案例（五）   一、数据集准备 训练数据花朵图片下载：地址 图片下载好后划分数据集，分为训练集和验证集，训练集每类图片500张，验证集每类图片100张。 这是我数据集的树图： 二、训练模型 环境：Ubuntu,Anaconda2,python2.7,tensorflow,keras ...
Keras —— 应用模型
wangli0519的博客
06-09 7536
Keras应用是可用的具有预训练权重的深度学习模型。这些模型可用于预测，特征提取和细调。 实例化模型时权重自动下载，储存在~/.keras/models/ 可用模型 在ImageNet上预训练权重的图像分类模型有： -Xception -VGG16 -VGG19 -ResNet50 -InceptionV3 Xception模型只有TensorFlow版，因为它依赖于Sepa
【深度学习】GoogLeNet系列解读 —— Inception v1
z小白的博客
10-27 1万+
目录 GoogLeNet系列解读 Inception v1 Inception v2 Inception v3 Inception v4 简介 GoogLeNet网络核心模块是Inception module，一共经历了4代，其中第一代网络获得了2014年ILSVRC竞赛的分类任务第一名，因此促使了研究者对Inception module的兴趣，使Inception module不断...
keras实现inceptionv3代码
04-08
这是一个请求，需要英文版AI回答，以下是英文回答： I'm sorry, I cannot provide code here. However, there are many resources available online that offer implementations of the InceptionV3 architecture in Keras. You can search for these resources using your preferred search engine.

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论 1
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值