自制数据集代码如下：
import tensorflow as tf
from PIL import Image
import numpy as np
import os
train_path = './mnist_image_label/mnist_train_jpg_60000/'
train_txt = './mnist_image_label/mnist_train_jpg_60000.txt'
x_train_savepath = './mnist_image_label/mnist_x_train.npy'
y_train_savepath = './mnist_image_label/mnist_y_train.npy'
test_path = './mnist_image_label/mnist_test_jpg_10000/'
test_txt = './mnist_image_label/mnist_test_jpg_10000.txt'
x_test_savepath = './mnist_image_label/mnist_x_test.npy'
y_test_savepath = './mnist_image_label/mnist_y_test.npy'
def generateds(path, txt):
f = open(txt, 'r') # 以只读形式打开txt文件
contents = f.readlines() # 读取文件中所有行
f.close() # 关闭txt文件
x, y_ = [], [] # 建立空列表
for content in contents: # 逐行取出
value = content.split() # 以空格分开，图片路径为value[0] , 标签为value[1] , 存入列表
img_path = path + value[0] # 拼出图片路径和文件名
img = Image.open(img_path) # 读入图片
img = np.array(img.convert('L')) # 图片变为8位宽灰度值的np.array格式
img = img / 255. # 数据归一化 （实现预处理）
x.append(img) # 归一化后的数据，贴到列表x
y_.append(value[1]) # 标签贴到列表y_
print('loading : ' + content) # 打印状态提示
x = np.array(x) # 变为np.array格式
y_ = np.array(y_) # 变为np.array格式
y_ = y_.astype(np.int64) # 变为64位整型
return x, y_ # 返回输入特征x，返回标签y_
if os.path.exists(x_train_savepath) and os.path.exists(y_train_savepath) and os.path.exists(
x_test_savepath) and os.path.exists(y_test_savepath):
print('-------------Load Datasets-----------------')
x_train_save = np.load(x_train_savepath)
y_train = np.load(y_train_savepath)
x_test_save = np.load(x_test_savepath)
y_test = np.load(y_test_savepath)
x_train = np.reshape(x_train_save, (len(x_train_save), 28, 28))
x_test = np.reshape(x_test_save, (len(x_test_save), 28, 28))
else:
print('-------------Generate Datasets-----------------')
x_train, y_train = generateds(train_path, train_txt)
x_test, y_test = generateds(test_path, test_txt)
print('-------------Save Datasets-----------------')
x_train_save = np.reshape(x_train, (len(x_train), -1))
x_test_save = np.reshape(x_test, (len(x_test), -1))
np.save(x_train_savepath, x_train_save)
np.save(y_train_savepath, y_train)
np.save(x_test_savepath, x_test_save)
np.save(y_test_savepath, y_test)
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['sparse_categorical_accuracy'])
model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1)
model.summary()
第一次运行时，由于没有进行数据集的训练，所以会先执行generateds函数，然后生成.npy格式的数据集。第二次运行时，由于第一次运行已经生成了数据集，系统会检测到数据集，进而执行数据的训练。
数据增强：
我们可能会因为拍照角度，阴影的种种原因导致数据的效果并不是那么好，所以可以引入数据增强。
image_gen_train.fit(x_train) 这里的fit需要输入一个四位数据，所以需要对x.train进行reshape处理。处理后可以把6万张，28行28列的数据 新增加一个通道为1的数据。这里的通道1指的是灰度值，之后就按照batch打包送入model.fit执行训练过程。
代码：
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1) # 给数据增加一个维度,从(60000, 28, 28)reshape为(60000, 28, 28, 1)
image_gen_train = ImageDataGenerator(
rescale=1. / 1., # 如为图像，分母为255时，可归至0～1
rotation_range=45, # 随机45度旋转
width_shift_range=.15, # 宽度偏移
height_shift_range=.15, # 高度偏移
horizontal_flip=False, # 水平翻转
zoom_range=0.5 # 将图像随机缩放阈量50％
)
image_gen_train.fit(x_train)
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['sparse_categorical_accuracy'])
model.fit(image_gen_train.flow(x_train, y_train, batch_size=32), epochs=5, validation_data=(x_test, y_test),
validation_freq=1)
model.summary()
断点续训很简单，就是可以保存训练的结果，使你在下次预测的时候不需要重新训练数据集，直接使用存放在本地的训练结果进行预测就行。
import tensorflow as tf
import os
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['sparse_categorical_accuracy'])
checkpoint_save_path = "./checkpoint/mnist.ckpt"
if os.path.exists(checkpoint_save_path + '.index'):
print('-------------load the model-----------------')
model.load_weights(checkpoint_save_path)
cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_save_path,
save_weights_only=True,
save_best_only=True)
history = model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1,
callbacks=[cp_callback])
model.summary()
save_weights_only 只保留模型参数
save_best_only 只保留最优模型
fit中加入callback，返回给history