Udacity优达学城 TensorFlow笔记2-对服饰图像分类

Fashion MNIST Dataset

有T恤、短靴等10个类别图像，每张图片为28*28像素的灰阶图像
共70000张图像，60K当作训练集，10K当作测试集

# !pip install -U tensorflow_datasets

# !pip install -U -i https://pypi.tuna.tsinghua.edu.cn/simple tensorflow-gpu==1.12.0 # 2.0.0a0

import tensorflow as tf
# 设置日志输出等级
tf.logging.set_verbosity(tf.logging.ERROR)
print(tf.__version__)
# tf.contrib 将被弃用，tf2.0 Eager Execution 变为默认执行模式
# import tensorflow.contrib.eager as tfe
# 开启Eager贪婪模式。一旦开启不能撤销！只执行一次。
tf.enable_eager_execution()
import tensorflow_datasets as tfds

import math
import numpy as np
import matplotlib.pyplot as plt

import tqdm
import tqdm.auto
tqdm.tqdm = tqdm.auto.tqdm

1.12.0

dataset, metadata = tfds.load('fashion_mnist', as_supervised=True, with_info=True)
train_dataset, test_dataset = dataset['train'], dataset['test']

class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
              'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']

num_train_examples = metadata.splits['train'].num_examples
num_test_examples = metadata.splits['test'].num_examples
print("Number of training examples: {}".format(num_train_examples))
print("Number of test examples: {}".format(num_test_examples))

Number of training examples: 60000
Number of test examples: 10000

# 标准化
def normalize(images, labels):
    images = tf.cast(images, tf.float32)
    images /= 255
    return images, labels

train_dataset = train_dataset.map(normalize)
test_dataset = test_dataset.map(normalize)

# 查看单个样本图像
for image, label in test_dataset.take(1):
    break
image = image.numpy().reshape((28, 28))
plt.figure()
plt.imshow(image, cmap=plt.cm.binary)
# 添加颜色标注
plt.colorbar()
# 去掉网格，默认也是去掉
plt.grid(False)

# 查看25个样本图像
plt.figure(figsize=(10, 10))
i = 0
for image, label in test_dataset.take(25):
    image = image.numpy().reshape((28, 28))
    plt.subplot(5, 5, i+1)
    plt.xticks([])
    plt.yticks([])
    plt.imshow(image, cmap=plt.cm.binary)
    plt.xlabel(class_names[label])
    i += 1

# 建立模型
model = tf.keras.Sequential([
    # 32个卷积输出，原来的一张图像变为32张卷积图像
    tf.keras.layers.Conv2D(32, (3,3), padding='same', activation=tf.nn.relu,
                           input_shape=(28, 28, 1)),
    # strides=2 步长为2
    tf.keras.layers.MaxPooling2D((2, 2), strides=2),
    tf.keras.layers.Conv2D(64, (3,3), padding='same', activation=tf.nn.relu),
    tf.keras.layers.MaxPooling2D((2, 2), strides=2),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation=tf.nn.relu),
    tf.keras.layers.Dense(10,  activation=tf.nn.softmax)
])

#model = tf.keras.Sequential([
#    # 扁平化Flatten 将二维图像转换为一维向量的过程
#    tf.keras.layers.Flatten(input_shape=(28, 28, 1)),
#    tf.keras.layers.Dense(128, activation=tf.nn.relu),
#    tf.keras.layers.Dense(10, activation=tf.nn.softmax)
#])

# 编译模型
model.compile(optimizer=tf.train.AdamOptimizer(0.001), 
              # 分类使用此损失函数
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

# 训练模型
BATCH_SIZE = 32
train_dataset = train_dataset.repeat().shuffle(num_train_examples).batch(BATCH_SIZE)
test_dataset = test_dataset.batch(BATCH_SIZE)

model.fit(train_dataset, epochs=10, steps_per_epoch=math.ceil(num_train_examples/BATCH_SIZE), verbose=1)

Epoch 1/10
1875/1875 [==============================] - 35s 19ms/step - loss: 0.4032 - acc: 0.8542
Epoch 2/10
1875/1875 [==============================] - 29s 16ms/step - loss: 0.2568 - acc: 0.9071
Epoch 3/10
1875/1875 [==============================] - 29s 15ms/step - loss: 0.2163 - acc: 0.9201
Epoch 4/10
1875/1875 [==============================] - 30s 16ms/step - loss: 0.1832 - acc: 0.9327
Epoch 5/10
1875/1875 [==============================] - 30s 16ms/step - loss: 0.1557 - acc: 0.9437
Epoch 6/10
1875/1875 [==============================] - 29s 16ms/step - loss: 0.1357 - acc: 0.9501
Epoch 7/10
1875/1875 [==============================] - 29s 16ms/step - loss: 0.1162 - acc: 0.9570
Epoch 8/10
1875/1875 [==============================] - 29s 16ms/step - loss: 0.0949 - acc: 0.9651
Epoch 9/10
1875/1875 [==============================] - 29s 16ms/step - loss: 0.0803 - acc: 0.9705
Epoch 10/10
1875/1875 [==============================] - 28s 15ms/step - loss: 0.0696 - acc: 0.9749

# 评估模型在测试集上的效果
test_loss, test_acc = model.evaluate(test_dataset, steps=math.ceil(num_test_examples/BATCH_SIZE))
print("Accuracy on test dataset:", test_acc)

313/313 [==============================] - 2s 7ms/step
Accuracy on test dataset: 0.9222

!jupyter nbconvert --to markdown "Udacity优达学城 TensorFlow笔记2-对服饰图像分类.ipynb"