摘要：最近在抽時(shí)間學(xué)習(xí)這個(gè)庫的使用，學(xué)的斷斷續(xù)續(xù)的，看官網(wǎng)上第一個(gè)案例就是訓(xùn)練手寫字符識(shí)別。此外，還需要有個(gè)，用來把訓(xùn)練的標(biāo)簽和實(shí)際的標(biāo)簽對(duì)應(yīng)，比如說對(duì)應(yīng)字母，對(duì)應(yīng)字母。然后使用的函數(shù)，把訓(xùn)練集和測(cè)試集分開。

最近在抽時(shí)間學(xué)習(xí)TensorFlow這個(gè)DL庫的使用，學(xué)的斷斷續(xù)續(xù)的，看官網(wǎng)上第一個(gè)案例就是訓(xùn)練手寫字符識(shí)別。我之前在做Weibo.cn驗(yàn)證碼識(shí)別的時(shí)候，自己搞了一個(gè)數(shù)據(jù)集，當(dāng)時(shí)用的c++庫tiny-dnn進(jìn)行訓(xùn)練的(見：驗(yàn)證碼破解技術(shù)四部曲之使用卷積神經(jīng)網(wǎng)絡(luò)（四）)，現(xiàn)在我把它移植到TensorFlow上試試。

完整代碼見：weibo.cn/tensorflow-impl

TensorFlow-1.0

scikit-learn-0.18

pillow

數(shù)據(jù)集下載地址：training_set.zip

解壓過后如下圖：

我把同一類的圖片放到了一個(gè)文件夾里，文件夾的名字也就是圖片的label，打開文件夾后可以看到字符的圖片信息。

下面，我們把數(shù)據(jù)加載到一個(gè)pickle文件里面，它需要有train_dataset、train_labels、test_dataset、test_labels四個(gè)變量代表訓(xùn)練集和測(cè)試集的數(shù)據(jù)和標(biāo)簽。

此外，還需要有個(gè)label_map，用來把訓(xùn)練的標(biāo)簽和實(shí)際的標(biāo)簽對(duì)應(yīng)，比如說3對(duì)應(yīng)字母M，4對(duì)應(yīng)字母N。

此部分的代碼見：load_models.py。注：很多的代碼參考自u(píng)dacity的deeplearning課程。

首先根據(jù)文件夾的來加載所有的數(shù)據(jù)，index代表訓(xùn)練里的標(biāo)簽，label代表實(shí)際的標(biāo)簽，使用PIL讀取圖片，并轉(zhuǎn)換成numpy數(shù)組。

import numpy as np
import os
from PIL import Image

def load_dataset():
    dataset = []
    labelset = []
    label_map = {}

    base_dir = "../trainer/training_set/"  # 數(shù)據(jù)集的位置
    labels = os.listdir(base_dir)

    for index, label in enumerate(labels):
        if label == "ERROR" or label == ".DS_Store":
            continue
        print "loading:", label, "index:", index
        try:
            image_files = os.listdir(base_dir + label)
            for image_file in image_files:
                image_path = base_dir + label + "/" + image_file
                im = Image.open(image_path).convert("L")
                dataset.append(np.asarray(im, dtype=np.float32))
                labelset.append(index)
            label_map[index] = label
        except: pass

    return np.array(dataset), np.array(labelset), label_map


dataset, labelset, label_map = load_dataset()

接下來，把數(shù)據(jù)打亂。

def randomize(dataset, labels):
    permutation = np.random.permutation(labels.shape[0])
    shuffled_dataset = dataset[permutation, :, :]
    shuffled_labels = labels[permutation]
    return shuffled_dataset, shuffled_labels


dataset, labelset = randomize(dataset, labelset)

然后使用scikit-learn的函數(shù)，把訓(xùn)練集和測(cè)試集分開。

from sklearn.model_selection import train_test_split
train_dataset, test_dataset, train_labels, test_labels = train_test_split(dataset, labelset)

在TensorFlow官網(wǎng)給的例子中，會(huì)把label進(jìn)行One-Hot Encoding，并把28*28的圖片轉(zhuǎn)換成了一維向量(784)。如下圖，查看官網(wǎng)例子的模型。

我也把數(shù)據(jù)轉(zhuǎn)換了一下，把32*32的圖片轉(zhuǎn)換成一維向量(1024)，并對(duì)標(biāo)簽進(jìn)行One-Hot Encoding。

def reformat(dataset, labels, image_size, num_labels):
    dataset = dataset.reshape((-1, image_size * image_size)).astype(np.float32)
    # Map 1 to [0.0, 1.0, 0.0 ...], 2 to [0.0, 0.0, 1.0 ...]
    labels = (np.arange(num_labels) == labels[:, None]).astype(np.float32)
    return dataset, labels

train_dataset, train_labels = reformat(train_dataset, train_labels, 32, len(label_map))
test_dataset, test_labels = reformat(test_dataset, test_labels, 32, len(label_map))
print "train_dataset:", train_dataset.shape
print "train_labels:", train_labels.shape
print "test_dataset:", test_dataset.shape
print "test_labels:", test_labels.shape

轉(zhuǎn)換后，格式就和minist一樣了。

最后，把數(shù)據(jù)保存到save.pickle里面。

save = {
    "train_dataset": train_dataset,
    "train_labels": train_labels,
    "test_dataset": test_dataset,
    "test_labels": test_labels,
    "label_map": label_map
}
with open("save.pickle", "wb") as f:
    pickle.dump(save, f)

加載完數(shù)據(jù)后，需要驗(yàn)證一下數(shù)據(jù)是否正確。我選擇的方法很簡(jiǎn)單，就是把trainset的第1個(gè)(或者第2個(gè)、第n個(gè))圖片打開，看看它的標(biāo)簽和看到的能不能對(duì)上。

import cPickle as pickle
from PIL import Image
import numpy as np

def check_dataset(dataset, labels, label_map, index):
    data = np.uint8(dataset[index]).reshape((32, 32))
    i = np.argwhere(labels[index] == 1)[0][0]
    im = Image.fromarray(data)
    im.show()
    print "label:", label_map[i]

if __name__ == "__main__":
    with open("save.pickle", "rb") as f:
        save = pickle.load(f)
        train_dataset = save["train_dataset"]
        train_labels = save["train_labels"]
        test_dataset = save["test_dataset"]
        test_labels = save["test_labels"]
        label_map = save["label_map"]

    # check if the image is corresponding to it"s label
    check_dataset(train_dataset, train_labels, label_map, 0)

運(yùn)行后，可以看到第一張圖片是Y，標(biāo)簽也是正確的。

數(shù)據(jù)加載好了之后，就可以開始訓(xùn)練了，訓(xùn)練的網(wǎng)絡(luò)就使用TensorFlow官網(wǎng)在Deep MNIST for Experts里提供的就好了。

此部分的代碼見：train.py。

先加載一下模型:

import cPickle as pickle
import numpy as np
import tensorflow as tf

with open("save.pickle", "rb") as f:
    save = pickle.load(f)
    train_dataset = save["train_dataset"]
    train_labels = save["train_labels"]
    test_dataset = save["test_dataset"]
    test_labels = save["test_labels"]
    label_map = save["label_map"]

image_size = 32
num_labels = len(label_map)

print "train_dataset:", train_dataset.shape
print "train_labels:", train_labels.shape
print "test_dataset:", test_dataset.shape
print "test_labels:", test_labels.shape
print "num_labels:", num_labels

minist的數(shù)據(jù)都是28*28的，把里面的網(wǎng)絡(luò)改完了之后，如下：

def weight_variable(shape):
    initial = tf.truncated_normal(shape, stddev=0.1)
    return tf.Variable(initial)


def bias_variable(shape):
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial)


def conv2d(x, W):
    return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding="SAME")


def max_pool_2x2(x):
    return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],
                          strides=[1, 2, 2, 1], padding="SAME")


graph = tf.Graph()
with graph.as_default():
    x = tf.placeholder(tf.float32, shape=[None, image_size * image_size])
    y_ = tf.placeholder(tf.float32, shape=[None, num_labels])

    x_image = tf.reshape(x, [-1, 32, 32, 1])

    # First Convolutional Layer
    W_conv1 = weight_variable([5, 5, 1, 32])
    b_conv1 = bias_variable([32])

    h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
    h_pool1 = max_pool_2x2(h_conv1)

    # Second Convolutional Layer
    W_conv2 = weight_variable([5, 5, 32, 64])
    b_conv2 = bias_variable([64])

    h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
    h_pool2 = max_pool_2x2(h_conv2)

    # Densely Connected Layer
    W_fc1 = weight_variable([image_size / 4 * image_size / 4 * 64, 1024])
    b_fc1 = bias_variable([1024])

    h_pool2_flat = tf.reshape(h_pool2, [-1, image_size / 4 * image_size / 4 * 64])
    h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)

    # Dropout
    keep_prob = tf.placeholder(tf.float32)
    h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

    # Readout Layer
    W_fc2 = weight_variable([1024, num_labels])
    b_fc2 = bias_variable([num_labels])

    y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2

    cross_entropy = tf.reduce_mean(
        tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))

    train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
    correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

主要改動(dòng)就是輸入層把28*28改成了image_size*image_size（32*32），然后第三層的全連接網(wǎng)絡(luò)把7*7改成了image_size/4*image_size/4（8*8），以及把10（手寫字符一共10類）改成了num_labels。

然后訓(xùn)練，我這里把batch_size改成了128，訓(xùn)練批次改少了。

batch_size = 128
with tf.Session(graph=graph) as session:
    tf.global_variables_initializer().run()
    print("Initialized")

    for step in range(2001):
        offset = (step * batch_size) % (train_labels.shape[0] - batch_size)
        # Generate a minibatch.
        batch_data = train_dataset[offset:(offset + batch_size), :]
        batch_labels = train_labels[offset:(offset + batch_size), :]

        if step % 50 == 0:
            train_accuracy = accuracy.eval(feed_dict={
                x: batch_data, y_: batch_labels, keep_prob: 1.0})
            test_accuracy = accuracy.eval(feed_dict={
                x: test_dataset, y_: test_labels, keep_prob: 1.0})
            print("Step %d, Training accuracy: %g, Test accuracy: %g" % (step, train_accuracy, test_accuracy))

        train_step.run(feed_dict={x: batch_data, y_: batch_labels, keep_prob: 0.5})

    print("Test accuracy: %g" % accuracy.eval(feed_dict={
        x: test_dataset, y_: test_labels, keep_prob: 1.0}))

運(yùn)行，可以看到識(shí)別率在不斷的上升。

最后，有了接近98%的識(shí)別率，只有4000個(gè)訓(xùn)練數(shù)據(jù)，感覺不錯(cuò)了。