def train(height = CAPTCHA_HEIGHT, width = CAPTCHA_WIDTH, y_size = len(CAPTCHA_LIST) * CAPTCHA_LEN):
acc_rate = 0.95
x = placeholder(float32, [None, height * width])
y = placeholder(float32, [None, y_size])
keep_prob = placeholder(float32)
y_conv = cnn_graph(x, keep_prob, (height, width))
optimizer = optimize_graph(y, y_conv)
accuracy = accuracy_graph(y, y_conv)
saver = Saver()
sess = Session()
sess.run(global_variables_initializer())
step = 0
while 1:
batch_x, batch_y = get_next_batch(64)
sess.run(optimizer, feed_dict = {x: batch_x, y: batch_y, keep_prob: 0.75})
if step % 100 == 0:
batch_x_test, batch_y_test = get_next_batch(100)
acc = sess.run(accuracy, feed_dict = {x: batch_x_test, y: batch_y_test, keep_prob: 1.0})
print(datetime.now().strftime('%c'), ' step:', step, ' accuracy:', acc)
if acc > acc_rate:
if not isdir('./model'):
mkdir('./model')
print('Saving to model/captcha.model')
saver.save(sess, './model/captcha.model', global_step = step)
print('Saved to model/captcha.model')
acc_rate += 0.005
if acc_rate >= 1:
break
step += 1
sess.close()
def train(height=CAPTCHA_HEIGHT,
width=CAPTCHA_WIDTH,
y_size=len(CAPTCHA_LIST) * CAPTCHA_LEN):
"""
cnn训练
:param height: 验证码高度
:param width: 验证码宽度
:param y_size: 验证码预备字符列表长度*验证码长度（默认为4）
:return:
"""
# cnn在图像大小是2的倍数时性能最高, 如果图像大小不是2的倍数，可以在图像边缘补无用像素
# 在图像上补2行，下补3行，左补2行，右补2行
# np.pad(image,((2,3),(2,2)), 'constant', constant_values=(255,))
acc_rate = 0.95 # 预设模型准确率标准
# 按照图片大小申请占位符
x = tf.placeholder(tf.float32, [None, height * width])
y = tf.placeholder(tf.float32, [None, y_size])
# 防止过拟合 训练时启用 测试时不启用 神经元使用率
keep_prob = tf.placeholder(tf.float32)
# cnn模型
y_conv = cnn_graph(x, keep_prob, (height, width))
# 优化
optimizer = optimize_graph(y, y_conv)
# 计算准确率
accuracy = accuracy_graph(y, y_conv)
# 启动会话.开始训练
saver = tf.train.Saver()
sess = tf.Session()
sess.run(tf.global_variables_initializer()) # 初始化
step = 0 # 步数
while 1:
batch_x, batch_y = get_next_batch(64)
sess.run(optimizer,
feed_dict={
x: batch_x,
y: batch_y,
keep_prob: 0.75
})
# 每训练一百次测试一次
if step % 100 == 0:
batch_x_test, batch_y_test = get_next_batch(100)
acc = sess.run(accuracy,
feed_dict={
x: batch_x_test,
y: batch_y_test,
keep_prob: 1.0
})
print(datetime.now().strftime('%c'), ' step:', step, ' accuracy:',
acc)
# 准确率满足要求，保存模型
if acc > acc_rate:
model_path = "./model/captcha.model"
saver.save(sess, model_path, global_step=step)
acc_rate += 0.01
if acc_rate > 0.99: # 准确率达到99%则退出
break
step += 1
sess.close()
def train(height=CAPTCHA_HEIGHT,
width=CAPTCHA_WIDTH,
y_size=len(CHARSET_LIST) * CAPTCHA_LEN):
"""
cnn训练
:param height: 验证码高度
:param width: 验证码宽度
:param y_size: 验证码预备字符列表长度*验证码长度（默认为4）
:return:
"""
target_accuracy = PRESET_ACCURACY # 预设模型准确率标准
# cnn 在图像大小是 2 的倍数时性能最高, 如果图像大小不是 2 的倍数，可以在图像边缘补无用像素
# 在图像上补 2 行，下补 3 行，左补 2 行，右补 2 行
# np.pad(image, ((2, 3), (2, 2)), 'constant', constant_values=(255,))
# 按照图片大小申请占位符
x = tf.compat.v1.placeholder(tf.float32, [None, height * width])
y = tf.compat.v1.placeholder(tf.float32, [None, y_size])
# 防止过拟合 训练时启用 测试时不启用 神经元使用率
keep_prob = tf.compat.v1.placeholder(tf.float32)
# cnn 模型
y_conv = cnn_graph(x, keep_prob, (height, width))
# 优化
optimizer = optimize_graph(y, y_conv)
# 计算准确率
accuracy = accuracy_graph(y, y_conv)
# 启动会话，开始训练
saver = tf.compat.v1.train.Saver()
sess = tf.compat.v1.Session()
sess.run(tf.compat.v1.global_variables_initializer()) # 初始化
step = 0 # 步数
while 1:
batch_x, batch_y = get_next_batch(64)
sess.run(optimizer,
feed_dict={
x: batch_x,
y: batch_y,
keep_prob: 0.75
})
# 每训练 100 次测试一次
if step % 100 == 0:
batch_x_test, batch_y_test = get_next_batch(100)
current_accuracy = sess.run(accuracy,
feed_dict={
x: batch_x_test,
y: batch_y_test,
keep_prob: 1.0
})
print(datetime.now().strftime('%c'), ' step:', step, ' accuracy:',
current_accuracy)
# 准确率满足要求，保存模型
if current_accuracy > target_accuracy:
model_path = os.path.join(MODEL_DIR, "captcha-model")
saver.save(sess, model_path, global_step=step)
target_accuracy += 0.01
if target_accuracy > FINAL_ACCURACY: # 准确率达到 FINAL_ACCURACY 则退出
break
step += 1
sess.close()
def train(height=CAPTCHA_HEIGHT, width=CAPTCHA_WIDTH, y_size=len(CAPTCHA_LIST)*CAPTCHA_LEN):
"""
cnn训练
:param height: 验证码高度
:param width: 验证码宽度
:param y_size: 验证码预备字符列表长度*验证码长度（默认为4）
:return:
"""
# cnn在图像大小是2的倍数时性能最高, 如果图像大小不是2的倍数，可以在图像边缘补无用像素
# 在图像上补2行，下补3行，左补2行，右补2行
# np.pad(image,((2,3),(2,2)), 'constant', constant_values=(255,))
acc_rate = 0.95 # 预设模型准确率标准
# 按照图片大小申请占位符
x = tf.placeholder(tf.float32, [None, height * width])
y = tf.placeholder(tf.float32, [None, y_size])
# 防止过拟合 训练时启用 测试时不启用 神经元使用率
keep_prob = tf.placeholder(tf.float32)
# cnn模型
y_conv = cnn_graph(x, keep_prob, (height, width))
# 优化
optimizer = optimize_graph(y, y_conv)
# 计算准确率
accuracy = accuracy_graph(y, y_conv)
# 启动会话.开始训练
saver = tf.train.Saver()
sess = tf.Session()
sess.run(tf.global_variables_initializer()) # 初始化
# 基于当前已有模型进行二次训练
if os.path.exists("model/checkpoint"):
model_file = tf.train.latest_checkpoint('model/')
saver.restore(sess, model_file)
step = 0 # 步数
while 1:
batch_x, batch_y = get_next_batch(64)
sess.run(optimizer, feed_dict={x: batch_x, y: batch_y, keep_prob: 0.75})
# 每训练一百次测试一次
if step % 100 == 0:
batch_x_test, batch_y_test = get_next_batch(100)
acc = sess.run(accuracy, feed_dict={x: batch_x_test, y: batch_y_test, keep_prob: 1.0})
print(datetime.now().strftime('%c'), ' step:', step, ' accuracy:', acc)
# 准确率满足要求，保存模型
if acc > acc_rate:
model_path = "./model/captcha.model"
saver.save(sess, model_path, global_step=step)
acc_rate += 0.01
if acc_rate > 0.99: # 准确率达到99%则退出
# 训练结束时保存pb模型
constant_graph = graph_util.convert_variables_to_constants(sess, sess.graph_def, ['output'])
with tf.gfile.FastGFile("./model/" + str(step) + 'model.pb', mode='wb') as f:
f.write(constant_graph.SerializeToString())
break
step += 1
sess.close()
summary = tf.summary.merge_all()
saver = tf.train.Saver()
# Initializer
init = tf.global_variables_initializer()
saver_step = 100
with tf.Session() as sess:
writer = tf.summary.FileWriter(logdir="./Tensorboard", graph=sess.graph)
sess.run(init)
for step in range(1, par.num_steps + 1):
# Get the next batch
batch_x, batch_y = util.get_next_batch(batch_size=par.batch_size,
image_size=par.image_size,
threshold=par.threshold)
batch_y = tf.one_hot(np.reshape(batch_y, [-1]), par.num_classes)
_, s = sess.run(
[training_op, summary],
feed_dict={
X: batch_x,
Y: sess.run(batch_y),
keep_prob: par.dropout,
global_step: step
})
writer.add_summary(s, step)
if step % par.display_step == 0 or step == 1:
loss, acc = sess.run([loss_op, accuracy],