level 1
a87871660
楼主
import os
import numpy as np
import tensorflow as tf
from pre import get_file, get_batch, valuation_batch
from nw import deep_CNN, losses, trainning, evaluation
ratio = 0.2
#keep_prob = 0.5
num_epochs = 1
n_class = 5
image_h = 640
image_w = 480
target_height = 790
target_width = 630
batch_size = 8
capacity = 256
max_step = 15000
learning_rate = 0.001
train_dir = 'E:\\DeepLearn\\picture_mosaic'
logs_train_dir = 'E:\\DeepLearn\\Classify_picture\\self\\'
train, train_label, valuation, valuation_label = get_file(train_dir, ratio)
train_batch, train_label_batch = get_batch(train, train_label, image_h, image_w, target_height, target_width, batch_size, capacity)
val_image_batch, val_label_batch = valuation_batch(valuation, valuation_label, image_h, image_w, target_height, target_width, batch_size, capacity)
is_training = tf.placeholder(tf.bool, shape=(), name='is_training')
x = tf.cond(is_training, lambda: train_batch, lambda: val_image_batch)
y = tf.cond(is_training, lambda: train_label_batch, lambda: val_label_batch)
train_logits = deep_CNN(x, batch_size, n_class)
#lish_train_logits = deep_CNN(val_image_batch, val_label_batch, n_class)
train_loss = losses(train_logits, y)
train_op = trainning(train_loss, learning_rate)
train_acc = evaluation(train_logits, y)
#可以将所有summary全部保存到磁盘，以便tensorboard显示。如果没有特殊要求，一般用这一句就可一显示训练时的各种信息了。
summary_op = tf.summary.merge_all()
sess = tf.Session()
# 指定一个文件用来保存图
train_writer = tf.summary.FileWriter(logs_train_dir, sess.graph)
saver = tf.train.Saver()
sess.run(tf.global_variables_initializer())
# 开启一个协调器
coord = tf.train.Coordinator()
# 使用start_queue_runners 启动队列填充
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
for step in np.arange(max_step):
if coord.should_stop():
#使用 coord.should_stop()来查询是否应该终止所有线程，当文件队列（queue）中的所有文件都已经读取出列的时候，会抛出一个 OutofRangeError 的异常，这时候就应该停止Sesson中的所有线程了;
break
_, tra_loss, tra_acc= sess.run([train_op, train_loss, train_acc], feed_dict={is_training: True})
if step % 100 == 0:
print('Step %d, train loss = %.2f, train accuracy = %.2f%%' % (step, tra_loss, tra_acc * 100.0))
#print('The learning rate is %s' %(_[1]))
val_loss, val_acc = sess.run([train_loss, train_acc], feed_dict={is_training: False})
print(val_loss, val_acc)
checkpoint_path = os.path.join(logs_train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
summary_str = sess.run(summary_op)
train_writer.add_summary(summary_str, step)
##调用train_writer的add_summary方法将训练过程以及训练步数保存
'''
if step % 2000 == 0 or (step + 1) == max_step:
checkpoint_path = os.path.join(logs_train_dir, 'model.ckpt')
# os.path.join连接两个或更多的路径名组件
#saver.save(sess, checkpoint_path, global_step=step)
summary.value.add(tag='evl_loss', simple_value=evl_loss)
summary.value.add(tag='evl_acc', simple_value=evl_acc)
saver.save(sess, checkpoint_path, global_step=step)
'''
except tf.errors.OutOfRangeError:
print('Done training -- epoch limit reached')
finally:
coord.request_stop()
coord.join(threads)
sess.close()
2019年07月21日 13点07分
1
import numpy as np
import tensorflow as tf
from pre import get_file, get_batch, valuation_batch
from nw import deep_CNN, losses, trainning, evaluation
ratio = 0.2
#keep_prob = 0.5
num_epochs = 1
n_class = 5
image_h = 640
image_w = 480
target_height = 790
target_width = 630
batch_size = 8
capacity = 256
max_step = 15000
learning_rate = 0.001
train_dir = 'E:\\DeepLearn\\picture_mosaic'
logs_train_dir = 'E:\\DeepLearn\\Classify_picture\\self\\'
train, train_label, valuation, valuation_label = get_file(train_dir, ratio)
train_batch, train_label_batch = get_batch(train, train_label, image_h, image_w, target_height, target_width, batch_size, capacity)
val_image_batch, val_label_batch = valuation_batch(valuation, valuation_label, image_h, image_w, target_height, target_width, batch_size, capacity)
is_training = tf.placeholder(tf.bool, shape=(), name='is_training')
x = tf.cond(is_training, lambda: train_batch, lambda: val_image_batch)
y = tf.cond(is_training, lambda: train_label_batch, lambda: val_label_batch)
train_logits = deep_CNN(x, batch_size, n_class)
#lish_train_logits = deep_CNN(val_image_batch, val_label_batch, n_class)
train_loss = losses(train_logits, y)
train_op = trainning(train_loss, learning_rate)
train_acc = evaluation(train_logits, y)
#可以将所有summary全部保存到磁盘，以便tensorboard显示。如果没有特殊要求，一般用这一句就可一显示训练时的各种信息了。
summary_op = tf.summary.merge_all()
sess = tf.Session()
# 指定一个文件用来保存图
train_writer = tf.summary.FileWriter(logs_train_dir, sess.graph)
saver = tf.train.Saver()
sess.run(tf.global_variables_initializer())
# 开启一个协调器
coord = tf.train.Coordinator()
# 使用start_queue_runners 启动队列填充
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
for step in np.arange(max_step):
if coord.should_stop():
#使用 coord.should_stop()来查询是否应该终止所有线程，当文件队列（queue）中的所有文件都已经读取出列的时候，会抛出一个 OutofRangeError 的异常，这时候就应该停止Sesson中的所有线程了;
break
_, tra_loss, tra_acc= sess.run([train_op, train_loss, train_acc], feed_dict={is_training: True})
if step % 100 == 0:
print('Step %d, train loss = %.2f, train accuracy = %.2f%%' % (step, tra_loss, tra_acc * 100.0))
#print('The learning rate is %s' %(_[1]))
val_loss, val_acc = sess.run([train_loss, train_acc], feed_dict={is_training: False})
print(val_loss, val_acc)
checkpoint_path = os.path.join(logs_train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
summary_str = sess.run(summary_op)
train_writer.add_summary(summary_str, step)
##调用train_writer的add_summary方法将训练过程以及训练步数保存
'''
if step % 2000 == 0 or (step + 1) == max_step:
checkpoint_path = os.path.join(logs_train_dir, 'model.ckpt')
# os.path.join连接两个或更多的路径名组件
#saver.save(sess, checkpoint_path, global_step=step)
summary.value.add(tag='evl_loss', simple_value=evl_loss)
summary.value.add(tag='evl_acc', simple_value=evl_acc)
saver.save(sess, checkpoint_path, global_step=step)
'''
except tf.errors.OutOfRangeError:
print('Done training -- epoch limit reached')
finally:
coord.request_stop()
coord.join(threads)
sess.close()
