def test_net():
args = parse_args()
config_reader = ConfigReader('experiments/configs/inception_v1.yml')
test_config = TestNetConfig(config_reader.get_test_config())
mode = 'RGB' if test_config.image_depth == 3 else 'L'
img = imread(args.im_path, mode=mode)
img = imresize(
img,
[test_config.image_height, test_config.image_width]) # height, width
k = 1 # select top k
with open(args.meta) as mf:
class_names = mf.read().splitlines()
class_names = list(filter(None, class_names))
net = InceptionV1(test_config)
logits = net.build_model()
values, indices = net.cal_result(logits)
ckpt_path = test_config.model_path
# start a session
saver = tf.train.Saver()
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
print('Model checkpoint path: {}'.format(ckpt_path))
try:
ckpt = tf.train.get_checkpoint_state(ckpt_path)
print('Restoring from {}...'.format(ckpt.model_checkpoint_path),
end=' ')
saver.restore(sess, ckpt.model_checkpoint_path)
print('done')
except FileNotFoundError:
raise 'Check your pretrained {:s}'.format(ckpt_path)
[prob, ind, out] = sess.run([values, indices, logits],
feed_dict={net.x: [img]})
prob = prob[0]
ind = ind[0]
print('Classification Result:')
for i in range(k):
print('Category Name: %s \nProbability: %.2f%%' %
(class_names[ind[i]], prob[i] * 100))
sess.close()
def train():
config_reader = ConfigReader('experiments/configs/vgg16.yml')
train_config = TrainNetConfig(config_reader.get_train_config())
data_config = DataConfig(config_reader.get_train_config())
train_log_dir = './logs/train/'
val_log_dir = './logs/val/'
if not os.path.exists(train_log_dir):
os.makedirs(train_log_dir)
if not os.path.exists(val_log_dir):
os.makedirs(val_log_dir)
net = VGG16(train_config)
# with tf.name_scope('input'):
# train_loader = CIFAR10BinDataLoader(data_config, is_train=True, is_shuffle=True)
# train_image_batch, train_label_batch = train_loader.generate_batch()
# val_loader = CIFAR10BinDataLoader(data_config, is_train=False, is_shuffle=False)
# val_image_batch, val_label_batch = val_loader.generate_batch()
train_op = net.build_model()
summaries = net.get_summary()
# saver = tf.train.Saver(tf.global_variables())
# summary_op = tf.summary.merge(summaries)
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
# net.load_with_skip(train_config.pre_train_weight, sess, ['fc6', 'fc7', 'fc8'])
# coord = tf.train.Coordinator()
# threads = tf.train.start_queue_runners(sess=sess, coord=coord)
with open("vgg.pbtxt", "w") as f:
f.write(str(sess.graph.as_graph_def()))
train_summary_writer = tf.summary.FileWriter(train_log_dir, sess.graph)
# val_summary_writer = tf.summary.FileWriter(val_log_dir, sess.graph)
sess.close()
return
def train():
config_reader = ConfigReader('experiments/configs/vgg16.yml')
train_config = TrainNetConfig(config_reader.get_train_config())
data_config = DataConfig(config_reader.get_train_config())
train_log_dir = './logs/train/'
val_log_dir = './logs/val/'
if not os.path.exists(train_log_dir):
os.makedirs(train_log_dir)
if not os.path.exists(val_log_dir):
os.makedirs(val_log_dir)
net = VGG16(train_config)
with tf.name_scope('input'):
train_loader = CIFAR10BinDataLoader(data_config,
is_train=True,
is_shuffle=True)
train_image_batch, train_label_batch = train_loader.generate_batch()
val_loader = CIFAR10BinDataLoader(data_config,
is_train=False,
is_shuffle=False)
val_image_batch, val_label_batch = val_loader.generate_batch()
train_op = net.build_model()
summaries = net.get_summary()
saver = tf.train.Saver(tf.global_variables())
summary_op = tf.summary.merge(summaries)
init = tf.global_variables_initializer()
sess = tf.Session()
# print(sess.graph.as_graph_def())
sess.run(init)
# print()
# net.load_with_skip(train_config.pre_train_weight, sess, ['fc6', 'fc7', 'fc8'])
coord = tf.compat.v1.train.Coordinator()
threads = tf.compat.v1.train.start_queue_runners(sess=sess, coord=coord)
train_summary_writer = tf.compat.v1.summary.FileWriter(
train_log_dir, sess.graph)
val_summary_writer = tf.compat.v1.summary.FileWriter(
val_log_dir, sess.graph)
vgg_watch_list = {
"conv1_1/relu:0": 0,
"conv2_1/relu:0": 1,
"conv3_1/relu:0": 2,
"conv4_1/relu:0": 3,
"conv5_1/relu:0": 4
}
conn = sqlite3.connect('./summary/test.db')
try:
for step in np.arange(train_config.max_step):
# print('step: ', step)
if coord.should_stop():
break
train_image, train_label = sess.run(
[train_image_batch, train_label_batch])
_, train_loss, train_acc = sess.run(
[train_op, net.loss, net.accuracy],
feed_dict={
net.x: train_image,
net.y: train_label
})
c = conn.cursor()
for item in vgg_watch_list:
data = sess.run(sess.graph.get_tensor_by_name(item),
feed_dict={
net.x: train_image,
net.y: train_label
})
c.execute(
"INSERT INTO LAYER_TABLE (STEP,NODE_ID,NODE_NAME,ACTIVATION_MIN,ACTIVATION_MAX, ACTIVATION_MEAN) VALUES (%d, %d, '%s', %.3f,%.3f,%.3f)"
% (step, vgg_watch_list[item], item, data.min(),
data.max(), data.mean()))
conn.commit()
val_image, val_label = sess.run([val_image_batch, val_label_batch])
# print(val_image.shape, val_label.shape)
# print(train_image.shape, train_label.shape)
val_loss, val_acc = sess.run([net.loss, net.accuracy],
feed_dict={
net.x: val_image,
net.y: val_label
})
c.execute(
"INSERT INTO LOSS_TABLE (STEP,TEST_LOSS,TEST_ACCURACY,TRAIN_LOSS,TRAIN_ACCURACY) VALUES (%d, %.3f,%.3f,%.3f,%.3f)"
% (step, val_loss, val_acc, train_loss, train_acc))
conn.commit()
if step % 50 == 0 or step + 1 == train_config.max_step:
print('===TRAIN===: Step: %d, loss: %.4f, accuracy: %.4f%%' %
(step, train_loss, train_acc))
summary_str = sess.run(summary_op,
feed_dict={
net.x: train_image,
net.y: train_label
})
train_summary_writer.add_summary(summary_str, step)
# if step % 200 == 0 or step + 1 == train_config.max_step:
# val_image, val_label = sess.run([val_image_batch, val_label_batch])
# plot_images = tf.summary.image('val_images_{}'.format(step % 200), val_image, 10)
# val_loss, val_acc, plot_summary = sess.run([net.loss, net.accuracy, plot_images], feed_dict={net.x: val_image, net.y: val_label})
# print('====VAL====: Step %d, val loss = %.4f, val accuracy = %.4f%%' % (step, val_loss, val_acc))
# summary_str = sess.run(summary_op, feed_dict={net.x: train_image, net.y: train_label})
# val_summary_writer.add_summary(summary_str, step)
# val_summary_writer.add_summary(plot_summary, step)
# if step % 2000 == 0 or step + 1 == train_config.max_step:
# checkpoint_path = os.path.join(train_log_dir, 'model.ckpt')
# saver.save(sess, checkpoint_path, global_step=step)
except tf.errors.OutOfRangeError:
print(
'===INFO====: Training completed, reaching the maximum number of steps'
)
finally:
coord.request_stop()
coord.join(threads)
sess.close()
def train():
config_reader = ConfigReader('experiments/configs/vgg16.yml')
train_config = TrainNetConfig(config_reader.get_train_config())
data_config = DataConfig(config_reader.get_train_config())
train_log_dir = './logs/train/'
val_log_dir = './logs/val/'
if not os.path.exists(train_log_dir):
os.makedirs(train_log_dir)
if not os.path.exists(val_log_dir):
os.makedirs(val_log_dir)
net = VGG16(train_config)
with tf.name_scope('input'):
train_loader = CIFAR10BinDataLoader(data_config,
is_train=True,
is_shuffle=True)
train_image_batch, train_label_batch = train_loader.generate_batch()
val_loader = CIFAR10BinDataLoader(data_config,
is_train=False,
is_shuffle=False)
val_image_batch, val_label_batch = val_loader.generate_batch()
train_op = net.build_model()
summaries = net.get_summary()
saver = tf.train.Saver(tf.global_variables())
summary_op = tf.summary.merge(summaries)
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
net.load_with_skip(train_config.pre_train_weight, sess,
['fc6', 'fc7', 'fc8'])
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
train_summary_writer = tf.summary.FileWriter(train_log_dir, sess.graph)
val_summary_writer = tf.summary.FileWriter(val_log_dir, sess.graph)
try:
for step in np.arange(train_config.max_step):
if coord.should_stop():
break
train_image, train_label = sess.run(
[train_image_batch, train_label_batch])
_, train_loss, train_acc = sess.run(
[train_op, net.loss, net.accuracy],
feed_dict={
net.x: train_image,
net.y: train_label
})
if step % 50 == 0 or step + 1 == train_config.max_step:
print('===TRAIN===: Step: %d, loss: %.4f, accuracy: %.4f%%' %
(step, train_loss, train_acc))
summary_str = sess.run(summary_op,
feed_dict={
net.x: train_image,
net.y: train_label
})
train_summary_writer.add_summary(summary_str, step)
if step % 200 == 0 or step + 1 == train_config.max_step:
val_image, val_label = sess.run(
[val_image_batch, val_label_batch])
plot_images = tf.summary.image(
'val_images_{}'.format(step % 200), val_image, 10)
val_loss, val_acc, plot_summary = sess.run(
[net.loss, net.accuracy, plot_images],
feed_dict={
net.x: val_image,
net.y: val_label
})
print(
'====VAL====: Step %d, val loss = %.4f, val accuracy = %.4f%%'
% (step, val_loss, val_acc))
summary_str = sess.run(summary_op,
feed_dict={
net.x: train_image,
net.y: train_label
})
val_summary_writer.add_summary(summary_str, step)
val_summary_writer.add_summary(plot_summary, step)
if step % 2000 == 0 or step + 1 == train_config.max_step:
checkpoint_path = os.path.join(train_log_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
except tf.errors.OutOfRangeError:
print(
'===INFO====: Training completed, reaching the maximum number of steps'
)
finally:
coord.request_stop()
coord.join(threads)
sess.close()
def train(_):
# Configure distibuted machines
ps_hosts = FLAGS.ps_hosts.split(",")
worker_hosts = FLAGS.worker_hosts.split(",")
cluster = tf.train.ClusterSpec({"ps": ps_hosts, "worker": worker_hosts})
# Create and start a server
server = tf.train.Server(cluster,
job_name=FLAGS.job_name,
task_index=FLAGS.task_index)
config_reader = ConfigReader('experiments/configs/inception_v1.yml')
train_config = TrainNetConfig(config_reader.get_train_config())
data_config = DataConfig(config_reader.get_train_config())
train_log_dir = './logs/train/'
val_log_dir = './logs/val/'
if not os.path.exists(train_log_dir):
os.makedirs(train_log_dir)
if not os.path.exists(val_log_dir):
os.makedirs(val_log_dir)
if FLAGS.job_name == "ps":
server.join()
else:
with tf.device(
tf.train.replica_device_setter(
worker_device="/job:worker/task:%d" % FLAGS.task_index,
cluster=cluster)):
net = InceptionV1(train_config, FLAGS.task_index)
with tf.name_scope('input'):
train_loader = Flowers102DataLoader(data_config,
is_train=True,
is_shuffle=True)
train_image_batch, train_label_batch = train_loader.generate_batch(
)
val_loader = Flowers102DataLoader(data_config,
is_train=False,
is_shuffle=False)
val_image_batch, val_label_batch = val_loader.generate_batch()
train_op = net.build_model()
summaries = net.get_summary()
saver = tf.train.Saver(tf.global_variables())
summary_op = tf.summary.merge(summaries)
init = tf.global_variables_initializer()
#sess = tf.Session()
sess = tf.train.MonitoredTrainingSession(
master=server.target,
is_chief=(FLAGS.task_index == 0),
hooks=net.sync_replicas_hook)
#sess.run(init)
#net.load_with_skip(train_config.pre_train_weight, sess, ['loss3_classifier'])
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
train_summary_writer = tf.summary.FileWriter(
train_log_dir, sess.graph)
val_summary_writer = tf.summary.FileWriter(val_log_dir, sess.graph)
step = 0
try:
#for step in np.arange(train_config.max_step):
while not sess.should_stop():
if coord.should_stop():
break
train_image, train_label = sess.run(
[train_image_batch, train_label_batch])
_, train_loss, train_acc = sess.run(
[train_op, net.loss, net.accuracy],
feed_dict={
net.x: train_image,
net.y: train_label
})
if step % (50 * train_config.update_delays
) == 0 or step + 1 == train_config.max_step:
print(
'===TRAIN===: Step: %d, loss: %.4f, accuracy: %.4f%%'
% (step, train_loss, train_acc))
summary_str = sess.run(summary_op,
feed_dict={
net.x: train_image,
net.y: train_label
})
train_summary_writer.add_summary(
summary_str, step / train_config.update_delays)
if step % (200 * train_config.update_delays
) == 0 or step + 1 == train_config.max_step:
pass
val_image, val_label = sess.run(
[val_image_batch, val_label_batch])
val_loss, val_acc = sess.run([net.loss, net.accuracy],
feed_dict={
net.x: val_image,
net.y: val_label
})
print(
'====VAL====: Step %d, val loss = %.4f, val accuracy = %.4f%%'
% (step, val_loss, val_acc))
summary_str = sess.run(summary_op,
feed_dict={
net.x: train_image,
net.y: train_label
})
val_summary_writer.add_summary(
summary_str, step / train_config.update_delays)
if step % 2000 == 0 or step + 1 == train_config.max_step:
checkpoint_path = os.path.join(train_log_dir,
'model.ckpt')
#saver.save(sess, checkpoint_path, global_step=step)
step += 1
except tf.errors.OutOfRangeError:
print(
'===INFO====: Training completed, reaching the maximum number of steps'
)
finally:
coord.request_stop()
coord.join(threads)
sess.close()