def train():
'''
Train CNN_tiny for a number of steps.
'''
with tf.Graph().as_default():
# globalなstep数
global_step = tf.Variable(0, trainable=False)
# 教師データ
image_input = ImageInput('./data/101Caltech_shuffles.txt')
print("the number of train data: %d" % (len(image_input.image_paths)))
images = tf.placeholder(tf.float32, [None, 224, 224, 3])
labels = tf.placeholder(tf.float32, [None, 101])
keep_conv = tf.placeholder(tf.float32)
keep_hidden = tf.placeholder(tf.float32)
# graphのoutput
logits, transform_result = model.inference(images, keep_conv,
keep_hidden)
# loss graphのoutputとlabelを利用
# loss = model.loss(logits, labels)
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits, labels))
predict_op = tf.argmax(logits, 1)
# 学習オペレーション
train_op = op.train(loss, global_step)
# サマリー
summary_op = tf.merge_all_summaries()
# 初期化オペレーション
init_op = tf.initialize_all_variables()
# Session
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=LOG_DEVICE_PLACEMENT))
# saver
#saver = tf.train.Saver(tf.all_variables())
sess.run(init_op)
# pretrainと全体を分けて保存
pretrain_params = {}
train_params = {}
for variable in tf.trainable_variables():
variable_name = variable.name
#print("parameter: %s" %(variable_name))
scope, name = variable_name.split("/")
target, _ = name.split(":")
if variable_name.find('spatial_transformer') < 0:
print("pretrain parameter: %s" % (variable_name))
pretrain_params[variable_name] = variable
print("train parameter: %s" % (variable_name))
train_params[variable_name] = variable
saver_cnn = tf.train.Saver(pretrain_params)
saver_transformers = tf.train.Saver(train_params)
# pretrained_model
if FLAGS.fine_tune:
ckpt = tf.train.get_checkpoint_state(PRETRAIN_DIR)
if ckpt and ckpt.model_checkpoint_path:
print("Pretrained Model Loading.")
saver_cnn.restore(sess, ckpt.model_checkpoint_path)
print("Pretrained Model Restored.")
else:
print("No Pretrained Model.")
# pretrained model from another type models.
# # saver
# print type(tf.all_variables())
# for variable in tf.trainable_variables():
# variable_name = variable.name
# variable_value = variable.eval(sess)
# if variable_name.find('softmax_linear') < 0 and variable_name.find('spatial_transformer') < 0:
# print("trained parameter: %s" %(variable_name))
# scope, name = variable_name.split("/")
# target, _ = name.split(":")
# with tf.variable_scope(scope, reuse=True):
# sess.run(tf.get_variable(target).assign(variable_value))
# trained_model = FLAGS.trained_model
# print trained_model
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
# サマリーのライターを設定
summary_writer = tf.train.SummaryWriter(TRAIN_DIR,
graph_def=sess.graph_def)
# max_stepまで繰り返し学習
for step in xrange(MAX_STEPS):
start_time = time.time()
previous_time = start_time
index = 0
batches = image_input.get_batches(FLAGS.batch_size)
for batch in batches:
train = batch[0]
label = batch[1]
_, loss_value = sess.run([train_op, loss],
feed_dict={
images: train,
labels: label,
keep_conv: 0.8,
keep_hidden: 0.5
})
if index % 10 == 0:
end_time = time.time()
duration = end_time - previous_time
num_examples_per_step = BATCH_SIZE * 10
examples_per_sec = num_examples_per_step / duration
print(
"%s: %d[epoch]: %d[iteration]: train loss %f: %d[examples/iteration]: %f[examples/sec]: %f[sec/iteration]"
% (datetime.now(), step, index, loss_value,
num_examples_per_step, examples_per_sec, duration))
index += 1
assert not np.isnan(
loss_value), 'Model diverged with loss = NaN'
# test_indices = np.arange(len(teX)) # Get A Test Batch
# np.random.shuffle(test_indices)
# test_indices = test_indices[0:5]
print "=" * 20
testx = train[0:2]
#print testx
testy = label[0:2]
print np.argmax(testy[0])
print np.argmax(testy[1])
output_vec, predict, cost_value = sess.run(
[logits, predict_op, loss],
feed_dict={
images: testx,
labels: testy,
keep_conv: 1.0,
keep_hidden: 1.0
})
print predict
print("test loss: %f" % (cost_value))
print "=" * 20
previous_time = end_time
index += 1
assert not np.isnan(
loss_value), 'Model diverged with loss = NaN'
# 100回ごと
if index % 100 == 0:
pass
summary_str = sess.run(summary_op,
feed_dict={
images: train,
labels: label,
keep_conv: 0.8,
keep_hidden: 0.5
})
# サマリーに書き込む
summary_writer.add_summary(summary_str, step)
if step % 1 == 0 or (step * 1) == MAX_STEPS:
pretrain_checkpoint_path = PRETRAIN_DIR + '/model.ckpt'
train_checkpoint_path = TRAIN_DIR + '/model.ckpt'
saver_cnn.save(sess,
pretrain_checkpoint_path,
global_step=step)
saver_transformers.save(sess,
train_checkpoint_path,
global_step=step)
coord.request_stop()
coord.join(threads)
sess.close()
def train():
'''
Train Network for a number of steps.
'''
with tf.Graph().as_default():
# globalなstep数
global_step = tf.Variable(0, trainable=False)
# 教師データ
if FLAGS.training_data_type == 0:
# directory feed inputs
image_input = ImageInput('caltech_samples_random.txt')
images = tf.placeholder(tf.float32,
shape=(FLAGS.batch_size, FLAGS.crop_size,
FLAGS.crop_size, FLAGS.image_depth))
labels = tf.placeholder(tf.int32, shape=(FLAGS.batch_size, ))
learning_rate_node = tf.placeholder(tf.float32, shape=[])
elif FLAGS.training_data_type == 1:
# tfrecords inputs
images, labels = data_inputs.distorted_inputs(TF_RECORDS)
# graphのoutput
# net, logits = model.inference(images)
logits = model.inference(images)
# loss graphのoutputとlabelを利用
loss = model.loss(logits, labels)
# 学習オペレーション
train_op = op.train(loss, global_step)
# バリデーション用 --> evalに移動するべき
# top_k_op = tf.nn.in_top_k(logits, labels, 1)
# saver
saver = tf.train.Saver(tf.all_variables())
# サマリー
summary_op = tf.merge_all_summaries()
# 初期化オペレーション
init_op = tf.initialize_all_variables()
# Session
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=LOG_DEVICE_PLACEMENT))
sess.run(init_op)
# 学習済みパラメータのロード
if FINE_TUNING:
print('fine tuning. trained_model: %s' % (FLAGS.trained_model))
#net.load(FLAGS.trained_model, sess)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
# サマリーのライターを設定
summary_writer = tf.train.SummaryWriter(TRAIN_DIR,
graph_def=sess.graph_def)
# max_stepまで繰り返し学習
for step in xrange(MAX_STEPS):
start_time = time.time()
# directory feed inputs
if FLAGS.training_data_type == 0:
train_images, train_labels = image_input.batches(
FLAGS.batch_size)
#print train_images
print train_labels
_, loss_value = sess.run(
[train_op, loss],
feed_dict={
images: train_images,
labels: train_labels,
learning_rate_node: FLAGS.learning_rate
})
elif FLAGS.training_data_type == 1:
# tfrecords inputs
#print kernel1.eval(session=sess)
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
# 10回ごと
if step % 10 == 0:
# stepごとの事例数 = mini batch size
num_examples_per_step = BATCH_SIZE
# 1秒ごとの事例数
examples_per_sec = num_examples_per_step / duration
# バッチごとの時間
sec_per_batch = float(duration)
# time, step数, loss, 1秒で実行できた事例数, バッチあたりの時間
format_str = '$s: step %d, loss = %.2f (%.1f examples/sec; %.3f sec/batch)'
print str(datetime.now()) + ': step' + str(
step) + ', loss= ' + str(loss_value) + ' ' + str(
examples_per_sec) + ' examples/sec; ' + str(
sec_per_batch) + ' sec/batch '
# 100回ごと
if step % 10 == 0:
if FLAGS.training_data_type == 0:
summary_str = sess.run(summary_op,
feed_dict={
images: train_images,
labels: train_labels
})
elif FLAGS.training_data_type == 1:
summary_str = sess.run(summary_op)
# サマリーに書き込む
#print sumary_str
summary_writer.add_summary(summary_str, step)
if step % 1000 == 0 or (step * 1) == MAX_STEPS:
checkpoint_path = TRAIN_DIR + '/model.ckpt'
saver.save(sess, checkpoint_path, global_step=step)
coord.request_stop()
coord.join(threads)
sess.close()
def train():
'''
Train CNN_tiny for a number of steps.
'''
with tf.Graph().as_default():
# globalなstep数
global_step = tf.Variable(0, trainable=False)
# 教師データ
image_input = ImageInput('./data/101Caltech_shuffles.txt')
print("the number of train data: %d" % (len(image_input.image_paths)))
images = tf.placeholder(tf.float32, [None, 224, 224, 3])
labels = tf.placeholder(tf.float32, [None, 101])
keep_conv = tf.placeholder(tf.float32)
keep_hidden = tf.placeholder(tf.float32)
# graphのoutput
logits, transform_result = model.inference(images, keep_conv, keep_hidden)
# loss graphのoutputとlabelを利用
# loss = model.loss(logits, labels)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, labels))
predict_op = tf.argmax(logits, 1)
# 学習オペレーション
train_op = op.train(loss, global_step)
# サマリー
summary_op = tf.merge_all_summaries()
# 初期化オペレーション
init_op = tf.initialize_all_variables()
# Session
sess = tf.Session(config=tf.ConfigProto(log_device_placement=LOG_DEVICE_PLACEMENT))
# saver
#saver = tf.train.Saver(tf.all_variables())
sess.run(init_op)
# pretrainと全体を分けて保存
pretrain_params = {}
train_params = {}
for variable in tf.trainable_variables():
variable_name = variable.name
#print("parameter: %s" %(variable_name))
scope, name = variable_name.split("/")
target, _ = name.split(":")
if variable_name.find('spatial_transformer') < 0:
print("pretrain parameter: %s" %(variable_name))
pretrain_params[variable_name] = variable
print("train parameter: %s" %(variable_name))
train_params[variable_name] = variable
saver_cnn = tf.train.Saver(pretrain_params)
saver_transformers = tf.train.Saver(train_params)
# pretrained_model
if FLAGS.fine_tune:
ckpt = tf.train.get_checkpoint_state(PRETRAIN_DIR)
if ckpt and ckpt.model_checkpoint_path:
print("Pretrained Model Loading.")
saver_cnn.restore(sess, ckpt.model_checkpoint_path)
print("Pretrained Model Restored.")
else:
print("No Pretrained Model.")
# pretrained model from another type models.
# # saver
# print type(tf.all_variables())
# for variable in tf.trainable_variables():
# variable_name = variable.name
# variable_value = variable.eval(sess)
# if variable_name.find('softmax_linear') < 0 and variable_name.find('spatial_transformer') < 0:
# print("trained parameter: %s" %(variable_name))
# scope, name = variable_name.split("/")
# target, _ = name.split(":")
# with tf.variable_scope(scope, reuse=True):
# sess.run(tf.get_variable(target).assign(variable_value))
# trained_model = FLAGS.trained_model
# print trained_model
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
# サマリーのライターを設定
summary_writer = tf.train.SummaryWriter(TRAIN_DIR, graph_def=sess.graph_def)
# max_stepまで繰り返し学習
for step in xrange(MAX_STEPS):
start_time = time.time()
previous_time = start_time
index = 0
batches = image_input.get_batches(FLAGS.batch_size)
for batch in batches:
train = batch[0]
label = batch[1]
_, loss_value = sess.run([train_op, loss], feed_dict={images: train, labels: label, keep_conv: 0.8, keep_hidden: 0.5})
if index % 10 == 0:
end_time = time.time()
duration = end_time - previous_time
num_examples_per_step = BATCH_SIZE * 10
examples_per_sec = num_examples_per_step / duration
print("%s: %d[epoch]: %d[iteration]: train loss %f: %d[examples/iteration]: %f[examples/sec]: %f[sec/iteration]" % (datetime.now(), step, index, loss_value, num_examples_per_step, examples_per_sec, duration))
index += 1
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
# test_indices = np.arange(len(teX)) # Get A Test Batch
# np.random.shuffle(test_indices)
# test_indices = test_indices[0:5]
print "="*20
testx = train[0:2]
#print testx
testy = label[0:2]
print np.argmax(testy[0])
print np.argmax(testy[1])
output_vec, predict, cost_value = sess.run([logits, predict_op, loss], feed_dict={images: testx, labels: testy, keep_conv: 1.0, keep_hidden: 1.0})
print predict
print("test loss: %f" % (cost_value))
print "="*20
previous_time = end_time
index += 1
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
# 100回ごと
if index % 100 == 0:
pass
summary_str = sess.run(summary_op, feed_dict={images: train, labels: label, keep_conv: 0.8, keep_hidden: 0.5})
# サマリーに書き込む
summary_writer.add_summary(summary_str, step)
if step % 1 == 0 or (step * 1) == MAX_STEPS:
pretrain_checkpoint_path = PRETRAIN_DIR + '/model.ckpt'
train_checkpoint_path = TRAIN_DIR + '/model.ckpt'
saver_cnn.save(sess, pretrain_checkpoint_path, global_step=step)
saver_transformers.save(sess, train_checkpoint_path, global_step=step)
coord.request_stop()
coord.join(threads)
sess.close()
def train():
'''
Train Network for a number of steps.
'''
with tf.Graph().as_default():
# globalなstep数
global_step = tf.Variable(0, trainable=False)
# 教師データ
if FLAGS.training_data_type == 0:
# directory feed inputs
image_input = ImageInput('caltech_samples_random.txt')
images = tf.placeholder(tf.float32, shape=(FLAGS.batch_size, FLAGS.crop_size, FLAGS.crop_size, FLAGS.image_depth))
labels = tf.placeholder(tf.int32, shape=(FLAGS.batch_size,))
learning_rate_node = tf.placeholder(tf.float32, shape=[])
elif FLAGS.training_data_type == 1:
# tfrecords inputs
images, labels = data_inputs.distorted_inputs(TF_RECORDS)
# graphのoutput
# net, logits = model.inference(images)
logits = model.inference(images)
# loss graphのoutputとlabelを利用
loss = model.loss(logits, labels)
# 学習オペレーション
train_op = op.train(loss, global_step)
# バリデーション用 --> evalに移動するべき
# top_k_op = tf.nn.in_top_k(logits, labels, 1)
# saver
saver = tf.train.Saver(tf.all_variables())
# サマリー
summary_op = tf.merge_all_summaries()
# 初期化オペレーション
init_op = tf.initialize_all_variables()
# Session
sess = tf.Session(config=tf.ConfigProto(log_device_placement=LOG_DEVICE_PLACEMENT))
sess.run(init_op)
# 学習済みパラメータのロード
if FINE_TUNING:
print('fine tuning. trained_model: %s' % (FLAGS.trained_model))
#net.load(FLAGS.trained_model, sess)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
# サマリーのライターを設定
summary_writer = tf.train.SummaryWriter(TRAIN_DIR, graph_def=sess.graph_def)
# max_stepまで繰り返し学習
for step in xrange(MAX_STEPS):
start_time = time.time()
# directory feed inputs
if FLAGS.training_data_type == 0:
train_images, train_labels = image_input.batches(FLAGS.batch_size)
#print train_images
print train_labels
_, loss_value = sess.run([train_op, loss], feed_dict={images: train_images, labels:train_labels, learning_rate_node: FLAGS.learning_rate})
elif FLAGS.training_data_type == 1:
# tfrecords inputs
#print kernel1.eval(session=sess)
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
# 10回ごと
if step % 10 == 0:
# stepごとの事例数 = mini batch size
num_examples_per_step = BATCH_SIZE
# 1秒ごとの事例数
examples_per_sec = num_examples_per_step / duration
# バッチごとの時間
sec_per_batch = float(duration)
# time, step数, loss, 1秒で実行できた事例数, バッチあたりの時間
format_str = '$s: step %d, loss = %.2f (%.1f examples/sec; %.3f sec/batch)'
print str(datetime.now()) + ': step' + str(step) + ', loss= '+ str(loss_value) + ' ' + str(examples_per_sec) + ' examples/sec; ' + str(sec_per_batch) + ' sec/batch '
# 100回ごと
if step % 10 == 0:
if FLAGS.training_data_type == 0:
summary_str = sess.run(summary_op, feed_dict={images: train_images, labels:train_labels})
elif FLAGS.training_data_type == 1:
summary_str = sess.run(summary_op)
# サマリーに書き込む
#print sumary_str
summary_writer.add_summary(summary_str, step)
if step % 1000 == 0 or (step * 1) == MAX_STEPS:
checkpoint_path = TRAIN_DIR + '/model.ckpt'
saver.save(sess, checkpoint_path, global_step=step)
coord.request_stop()
coord.join(threads)
sess.close()
