Kaggle-VGG16

程序员文章站 2022-07-12 08:14:17

...

一、导包准备：

from keras import models
from keras.layers import Conv2D, MaxPooling2D, Dense, BatchNormalization, GlobalAveragePooling2D, Dropout

import numpy as np
import cv2
from sklearn.model_selection import train_test_split

import matplotlib.pyplot as plt

二、加载数据：

n = 25000
width = 128

x = np.zeros(shape=(n, width, width, 3), dtype=np.uint8)
y = np.zeros(shape=(n, 1), dtype=np.uint8)

for i in range(12500):
    '''
    image = cv2.imread('E:/09-KAGGLE/dogs-vs-cats-redux-kernels-edition/train/train/cat.%d.jpg' % i)
    try:
        if image is None:
            print('error: train/cat.%d.jpg' % i)
            break;
    except AttributeError:
        pass
    '''
    x[i] = cv2.resize(cv2.imread('E:/09-KAGGLE/dogs-vs-cats-redux-kernels-edition/train/train/cat.%d.jpg' % i), dsize=(width, width))
    x[i+12500] = cv2.resize(cv2.imread('E:/09-KAGGLE/dogs-vs-cats-redux-kernels-edition/train/train/dog.%d.jpg' % i), dsize=(width, width))
y[12500:] = 1

原来相对路径读取文件的时候不定时报错：
error: (-215:Assertion failed) !ssize.empty() in function 'cv::resize’
原因：图像损坏无法读取，重新解压文件。
文件地址：https://pan.baidu.com/s/1H-aPZ5fOJT21lh7aIK4X4g
（网盘深度学习文件夹下dogs-vs-cats-redux-kernels-edition压缩包）
Kaggle-VGG16

解决办法参考：添加链接描述
Kaggle-VGG16

三、训练集与测试集拆分：

x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2)

四、搭建模型：

model = models.Sequential()

model.add(Conv2D(filters=32,kernel_size=(3,3), activation='relu', input_shape=(128,128,3), padding='same'))
model.add(BatchNormalization())


model.add(Conv2D(filters=32,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2,2)))


model.add(Conv2D(filters=64,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())

model.add(Conv2D(filters=64,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2,2)))



model.add(Conv2D(filters=128,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())

model.add(Conv2D(filters=128,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())


model.add(Conv2D(filters=128,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2,2)))

model.add(Conv2D(filters=256,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())


model.add(Conv2D(filters=256,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())

model.add(Conv2D(filters=256,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2,2)))



model.add(Conv2D(filters=512,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())


model.add(Conv2D(filters=512,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())


model.add(Conv2D(filters=512,kernel_size=(3,3), activation='relu', padding='same'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2,2)))


model.add(GlobalAveragePooling2D())


model.add(Dropout(0.5))

model.add(Dense(units=1, activation='sigmoid'))

这模型搭了我一晚上！每一层卷积的时候，如果不设定padding操作‘same’，图像会越来越小，后面的会报错：
Negative dimension size caused by subtracting 3 from 2 for ‘Conv2D’。
一层一层的summary后发现卷积到其中某一层时特征图大小只有2×2大小，卷积核大小是3×3，所以会有什么3 from 2的错误。
Conv2D层声明padding = 'same’就解决了。

五、模型编译：

model.compile(optimizer = 'adam', loss='binary_crossentropy', metrics=['accuracy'])

为什么要编译模型？
《Python深度学习》（弗朗索瓦·肖莱著张亮译人民邮电出版社）Page22:
Kaggle-VGG16
六、模型训练：

h = model.fit(x_train,y_train, batch_size=128, epochs=50, validation_data=(x_test, y_test))

七、训练结果可视化：

plt.figure(figsize=(10,4))
plt.subplot(1, 2, 1)
plt.plot(h.history['loss'])
plt.plot(h.history['val_loss'])
plt.legend(['loss', 'val_loss'])
plt.ylabel('loss')
plt.xlabel('epoch')

print(h.history)

plt.subplot(1, 2, 2)
plt.plot(h.history['accuracy'])
plt.plot(h.history['val_accuracy'])
plt.legend(['acc', 'val_cc'])
plt.ylabel('acc')
plt.xlabel('epoch')

Kaggle-VGG16
八、保存模型：

model.save('cats_and_dogs_small_2.h5')

九、model.summary
model.summary()
Model: “sequential_1”

Layer (type) Output Shape Param #

conv2d_1 (Conv2D) (None, 128, 128, 32) 896

batch_normalization_1 (Batch (None, 128, 128, 32) 128

conv2d_2 (Conv2D) (None, 128, 128, 32) 9248

batch_normalization_2 (Batch (None, 128, 128, 32) 128

max_pooling2d_1 (MaxPooling2 (None, 64, 64, 32) 0

conv2d_3 (Conv2D) (None, 64, 64, 64) 18496

batch_normalization_3 (Batch (None, 64, 64, 64) 256

conv2d_4 (Conv2D) (None, 64, 64, 64) 36928

batch_normalization_4 (Batch (None, 64, 64, 64) 256

max_pooling2d_2 (MaxPooling2 (None, 32, 32, 64) 0

conv2d_5 (Conv2D) (None, 32, 32, 128) 73856

batch_normalization_5 (Batch (None, 32, 32, 128) 512

conv2d_6 (Conv2D) (None, 32, 32, 128) 147584

batch_normalization_6 (Batch (None, 32, 32, 128) 512

conv2d_7 (Conv2D) (None, 32, 32, 128) 147584

batch_normalization_7 (Batch (None, 32, 32, 128) 512

max_pooling2d_3 (MaxPooling2 (None, 16, 16, 128) 0

conv2d_8 (Conv2D) (None, 16, 16, 256) 295168

batch_normalization_8 (Batch (None, 16, 16, 256) 1024

conv2d_9 (Conv2D) (None, 16, 16, 256) 590080

batch_normalization_9 (Batch (None, 16, 16, 256) 1024

conv2d_10 (Conv2D) (None, 16, 16, 256) 590080

batch_normalization_10 (Batc (None, 16, 16, 256) 1024

max_pooling2d_4 (MaxPooling2 (None, 8, 8, 256) 0

conv2d_11 (Conv2D) (None, 8, 8, 512) 1180160

batch_normalization_11 (Batc (None, 8, 8, 512) 2048

conv2d_12 (Conv2D) (None, 8, 8, 512) 2359808

batch_normalization_12 (Batc (None, 8, 8, 512) 2048

conv2d_13 (Conv2D) (None, 8, 8, 512) 2359808

batch_normalization_13 (Batc (None, 8, 8, 512) 2048

max_pooling2d_5 (MaxPooling2 (None, 4, 4, 512) 0

global_average_pooling2d_1 ( (None, 512) 0

dropout_1 (Dropout) (None, 512) 0

dense_1 (Dense) (None, 1) 513

Total params: 7,821,729
Trainable params: 7,815,969
Non-trainable params: 5,760

十、plot_model
Kaggle-VGG16