def tower_loss(scope, keep_prob):
images, labels = captcha.inputs(train=True, batch_size=FLAGS.batch_size)
logits = captcha.inference(images, keep_prob)
_ = captcha.loss(logits, labels)
losses = tf.get_collection('losses', scope)
total_loss = tf.add_n(losses, name='total_loss')
return total_loss
def run_train():
"""Train CAPTCHA for a number of steps."""
with tf.Graph().as_default():
images, labels = captcha.inputs(train=True, batch_size=192)
logits = captcha.inference(images, keep_prob=0.5)
loss = captcha.loss(logits, labels)
train_op = captcha.training(loss)
saver = tf.compat.v1.train.Saver()
init_op = tf.group(tf.compat.v1.global_variables_initializer(),
tf.compat.v1.local_variables_initializer())
sess = tf.compat.v1.Session()
#sess.run(init_op)
try:
#saver.restore(sess, FLAGS.checkpoint)
saver.restore(sess,
tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
except Exception as e:
print(e)
sess.run(init_op)
#exit()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
_loss = 100000
step = 0
while not coord.should_stop():
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
if step % 10 == 0:
print('>> Step %d run_train: loss = %f (%.4f sec)' %
(step, loss_value, duration))
if _loss > loss_value:
print('>> %s STEP %d LOSS %f SPAN %.4f' %
(datetime.now(), step, loss_value, duration))
saver.save(sess, FLAGS.checkpoint, global_step=step)
_loss = loss_value
#open('learning.log', 'a').write(step.__str__()+'\t'+loss_value.__str__()+'\n')
step += 1
except Exception as e:
#print('>> %s STEP:' % (datetime.now()))
saver.save(sess, FLAGS.checkpoint, global_step=step)
coord.request_stop(e)
finally:
coord.request_stop()
coord.join(threads)
sess.close()
def run_train():
"""Train CAPTCHA for a number of steps."""
with tf.Graph().as_default():
images, labels = captcha.inputs(
train=True, batch_size=FLAGS.batch_size)
logits = captcha.inference(images, keep_prob=1)
loss = captcha.loss(logits, labels)
train_op = captcha.training(loss)
saver = tf.train.Saver(tf.global_variables())
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
with tf.Session() as sess:
sess.run(init_op)
writer = tf.summary.FileWriter('./logs/', sess.graph)
# load DATA
if hasCheckpoint():
saver.restore(
sess, tf.train.latest_checkpoint('./captcha_train'))
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
step = 0
while not coord.should_stop():
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
if step % 10 == 0:
print('>> Step %d run_train: loss = %.2f (%.3f sec)' % (step, loss_value,
duration))
if step % 100 == 0:
print('>> %s Saving in %s' %
(datetime.now(), FLAGS.checkpoint))
saver.save(sess, FLAGS.checkpoint, global_step=step)
step += 1
except Exception as e:
print('>> %s Saving in %s' %
(datetime.now(), FLAGS.checkpoint))
saver.save(sess, FLAGS.checkpoint, global_step=step)
coord.request_stop(e)
finally:
coord.request_stop()
coord.join(threads)
writer.close()
def run_train():
"""Train CAPTCHA for a number of steps."""
with tf.Graph().as_default():
images, labels = captcha.inputs(train=True,
batch_size=FLAGS.batch_size)
logits = captcha.inference(images, keep_prob=0.5)
loss = captcha.loss(logits, labels)
train_op = captcha.training(loss)
saver = tf.train.Saver(tf.global_variables())
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess = tf.Session()
sess.run(init_op)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
checkpoint_path = os.path.join(FLAGS.model_dir, 'checkpoint')
print("checkpoint_path: %s" % checkpoint_path)
try:
step = 0
while not coord.should_stop():
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
if step % 10 == 0:
print('>> Step %d run_train: loss = %.2f (%.3f sec)' %
(step, loss_value, duration))
if step % 100 == 0:
print('>> %s Saving in %s' %
(datetime.now(), checkpoint_path))
saver.save(sess, checkpoint_path, global_step=step)
step += 1
except Exception as e:
print('>> %s Saving in %s' % (datetime.now(), checkpoint_path))
saver.save(sess, checkpoint_path, global_step=step)
coord.request_stop(e)
finally:
coord.request_stop()
coord.join(threads)
sess.close()
def main():
output_types, output_shapes, train_handle = captcha_data.generate_handle(
"tfrecords/train.tfrecords")
_, _, validation_handel = captcha_data.generate_handle(
"tfrecords/validation.tfrecords")
handle = tf.placeholder(tf.string, shape=[])
iterator = tf.data.Iterator.from_string_handle(string_handle=handle,
output_types=output_types,
output_shapes=output_shapes)
next_element = iterator.get_next()
images = next_element['image']
labels = next_element['label']
logits = captcha_model.inference(images)
loss = captcha_model.loss(logits, labels)
accuracy = captcha_model.evaluation(logits, labels)
train_op = captcha_model.train(loss)
init = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
config = tf.ConfigProto(allow_soft_placement=True)
sess = tf.Session(config=config)
sess.run(init)
train_handle_value = sess.run(train_handle)
validation_handel_value = sess.run(validation_handel)
for i in range(10000):
_, loss_value = sess.run([train_op, loss],
feed_dict={handle: train_handle_value})
if (i + 1) % 10 == 0:
accuracy_value = sess.run(
accuracy, feed_dict={handle: validation_handel_value})
print("loop: %d, loss: %f accuracy: %f" %
(i + 1, loss_value, accuracy_value))
saver = tf.train.Saver()
saver.save(sess, "save/model")
sess.close()
def run_train():
"""Train CAPTCHA for a number of steps."""
with tf.Graph().as_default():
images, labels = captcha.inputs(train=True,
batch_size=FLAGS.batch_size)
test_images, test_labels = captcha.inputs(train=False,
batch_size=FLAGS.batch_size)
logits = captcha.inference(images, keep_prob=0.7, is_training=True)
test_logits = captcha.inference(test_images,
keep_prob=1,
is_training=False)
loss = captcha.loss(logits, labels)
correct = captcha.evaluation(logits, labels)
test_correct = captcha.evaluation(test_logits, test_labels)
eval_correct = correct / FLAGS.batch_size
tf.summary.scalar('precision', eval_correct)
tf.summary.scalar('loss', loss)
tf.summary.image('image', images, 10)
summary = tf.summary.merge_all()
train_op = captcha.training(loss)
saver = tf.train.Saver(tf.global_variables())
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
with tf.Session() as sess:
sess.run(init_op)
summary_writer = tf.summary.FileWriter(FLAGS.train_dir, sess.graph)
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.train_dir))
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
step = 0
while not coord.should_stop():
start_time = time.time()
_, loss_value, test_co, train_co = sess.run(
[train_op, loss, test_correct, correct])
result = sess.run(summary) #merged也是需要run的
summary_writer.add_summary(result, step)
summary_writer.flush()
duration = time.time() - start_time
if step % 100 == 0:
print(
'>> Step %d run_train: loss = %.2f , train_correct= %d, test_correct= %d (%.3f sec)'
% (step, loss_value, train_co, test_co, duration))
with open('test.txt', 'a') as f:
f.write(str(test_co) + '\n')
if step % 300 == 0:
print('>> %s Saving in %s' %
(datetime.now(), FLAGS.checkpoint))
saver.save(sess, FLAGS.checkpoint, global_step=step)
step += 1
if step > 200000:
break
except Exception as e:
print('>> %s Saving in %s' %
(datetime.now(), FLAGS.checkpoint))
saver.save(sess, FLAGS.checkpoint, global_step=step)
coord.request_stop(e)
finally:
coord.request_stop()
coord.join(threads)
def run_train():
"""Train CAPTCHA for a number of steps."""
with tf.Graph().as_default():
images, labels = captcha.inputs(train=True,
batch_size=FLAGS.batch_size)
test_images, test_labels = captcha.inputs(train=False,
batch_size=FLAGS.batch_size)
logits = captcha.inference(images, keep_prob=0.75, is_training=True)
test_logits = captcha.inference(test_images,
keep_prob=1,
is_training=False)
loss = captcha.loss(logits, labels)
test_correct = captcha.evaluation(test_logits, test_labels) #test
correct = captcha.evaluation(logits, labels) #train
train_precision = correct / FLAGS.batch_size
tf.summary.scalar('loss', loss)
tf.summary.scalar('train_precision', train_precision)
tf.summary.image('images', images, 10)
summary = tf.summary.merge_all()
train_op = captcha.training(loss)
saver = tf.train.Saver()
# init_op = tf.group(tf.global_variables_initializer(),
# tf.local_variables_initializer())
sess = tf.Session()
summary_writer = tf.summary.FileWriter(FLAGS.train_dir, sess.graph)
# sess.run(init_op)
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
step = 120140
while not coord.should_stop():
start_time = time.time()
_, loss_value, test_value, train_value = sess.run(
[train_op, loss, test_correct, correct])
summary_str = sess.run(summary)
summary_writer.add_summary(summary_str, step)
summary_writer.flush()
duration = time.time() - start_time
step += 1
if step % 10 == 0:
print(
'>> Step %d run_train: loss = %.2f, test = %.2f, train = %.2f (%.3f sec)'
%
(step, loss_value, test_value, train_value, duration))
#-------------------------------
if step % 100 == 0:
print('>> %s Saving in %s' %
(datetime.now(), FLAGS.checkpoint))
saver.save(sess, FLAGS.checkpoint, global_step=step)
print(images.shape.as_list(), labels.shape.as_list())
if step > 2000000:
break
except KeyboardInterrupt:
print('INTERRUPTED')
coord.request_stop()
except Exception as e:
coord.request_stop(e)
finally:
saver.save(sess, FLAGS.checkpoint, global_step=step)
print('Model saved in file :%s' % FLAGS.checkpoint)
coord.request_stop()
coord.join(threads)
sess.close()
def main():
filename_queue = tf.train.string_input_producer(
["../images/tfrecords/train.tfrecords"])
image, label = captcha_inputs.read_records(filename_queue)
images, labels = captcha_inputs.records_inputs(image, label,
MIN_AFTER_DEQUEUE)
validation_filename_queue = tf.train.string_input_producer(
["../images/tfrecords/validation.tfrecords"])
validation_image, validation_label = captcha_inputs.read_records(
validation_filename_queue)
validation_images, validation_labels = captcha_inputs.records_inputs(
validation_image, validation_label, VALIDATION_MIN_AFTER_DEQUEUE)
# images, labels = captcha_inputs.inputs("../images/train", MIN_AFTER_DEQUEUE)
# validation_images, validation_labels = captcha_inputs.inputs("../images/validation", VALIDATION_MIN_AFTER_DEQUEUE)
# images, labels = captcha_inputs.inputs("/home/windows98/TensorFlow/application/CAPTCHA/images/"
# , MIN_AFTER_DEQUEUE)
# eval_images, eval_labels = captcha_inputs("/home/windows98/TensorFlow/application/CAPTCHA/eval_images/PNG",
# 5, 100, 300)
# TODO: add validation
logits = captcha_model.inference(images)
tf.get_variable_scope().reuse_variables()
with tf.device("/cpu:0"):
validation_logits = captcha_model.inference(validation_images)
accuracy = captcha_model.evaluation(validation_logits,
validation_labels)
loss = captcha_model.loss(logits, labels)
train_op = captcha_model.train(loss)
init = tf.group(tf.initialize_all_variables(),
tf.initialize_local_variables())
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
sess.run(init)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
start_time = datetime.now()
try:
index = 1
while not coord.should_stop():
_, loss_value = sess.run([train_op, loss])
print("step: " + str(index) + " loss:" + str(loss_value))
if index % 5 == 0:
validation_accuracy = sess.run(accuracy)
print("validation accuracy: " + str(validation_accuracy))
index += 1
except tf.errors.OutOfRangeError:
print('Done training -- epoch limit reached')
end_time = datetime.now()
print("Time Consumption: " + str(end_time - start_time))
except KeyboardInterrupt:
print("keyboard interrupt detected, stop running")
del sess
finally:
# When done, ask the threads to stop.
coord.request_stop()
# Wait for threads to finish.
coord.join(threads)
sess.close()
del sess