def __init__(self,
checkpoint_path,
logits,
im_shape,
num_channels,
ind_to_label,
moving_average_decay,
batch_size=1):
"""
checkpoint_path: path to model checkpoint file
logits is a function taking a tensor -> float array
logits must take a single tensor argument
ind_to_label is object such that ind_to_label[ind] = label
"""
self.im_shape = im_shape
self.num_channels = num_channels
self.batch_size = batch_size
self.g = tf.Graph()
with self.g.as_default():
self.sess = tf.Session()
self.checkpoint_step = checkpoint_path.split('-')[-1]
self.checkpoint_path = checkpoint_path
self.images_placeholder = tf.placeholder(
"float", shape=[im_shape[0], im_shape[1], num_channels])
self.logits = logits(self.images_placeholder)
self.saver = utils.get_saver(moving_average_decay,
nontrainable_restore_names="bn_")
self.ind_to_label = ind_to_label
# restore the model's parameters
self.saver.restore(self.sess, checkpoint_path)
def evaluate():
"""Eval captcha model for a number of steps."""
with tf.Graph().as_default():
# Get images and labels for captchas
eval_data = FLAGS.eval_data == 'test'
images, all_labels = inputs(eval_data=eval_data)
# split the labels
print(images)
labels = tf.split(1, 6, all_labels)
# Build a Graph that computes the logits predictions from the
# inference model.
logits, _ = inference_captcha_mean_subtracted(images, False, False)
print(logits[0])
print(labels[0])
ls = [tf.reshape(label, [FLAGS.batch_size]) for label in labels]
print(ls)
# Calculate predictions.
top_k_ops = [
tf.nn.in_top_k(logit, label, 1)
for logit, label in zip(logits, ls)
]
# Restore the moving average version of the learned variables for eval.
saver = get_saver(MOVING_AVERAGE_DECAY, "bn_")
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
graph_def = tf.get_default_graph().as_graph_def()
summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir,
graph_def=graph_def)
while True:
eval_once(saver, summary_writer, top_k_ops, summary_op)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
def __init__(self, checkpoint_path, logits, im_shape, num_channels, ind_to_label, moving_average_decay, batch_size=1):
"""
checkpoint_path: path to model checkpoint file
logits is a function taking a tensor -> float array
logits must take a single tensor argument
ind_to_label is object such that ind_to_label[ind] = label
"""
self.im_shape = im_shape
self.num_channels = num_channels
self.batch_size = batch_size
self.g = tf.Graph()
with self.g.as_default():
self.sess = tf.Session()
self.checkpoint_step = checkpoint_path.split('-')[-1]
self.checkpoint_path = checkpoint_path
self.images_placeholder = tf.placeholder("float", shape=[im_shape[0], im_shape[1], num_channels])
self.logits = logits(self.images_placeholder)
self.saver = utils.get_saver(moving_average_decay, nontrainable_restore_names="bn_")
self.ind_to_label = ind_to_label
# restore the model's parameters
self.saver.restore(self.sess, checkpoint_path)
def evaluate():
"""Eval captcha model for a number of steps."""
with tf.Graph().as_default():
# Get images and labels for captchas
eval_data = FLAGS.eval_data == 'test'
images, all_labels = inputs(eval_data=eval_data)
# split the labels
print(images)
labels = tf.split(1, 6, all_labels)
# Build a Graph that computes the logits predictions from the
# inference model.
logits, _ = inference_captcha_mean_subtracted(images, False, False)
print(logits[0])
print(labels[0])
ls = [tf.reshape(label, [FLAGS.batch_size]) for label in labels]
print(ls)
# Calculate predictions.
top_k_ops = [tf.nn.in_top_k(logit, label, 1) for logit, label in zip(logits, ls)]
# Restore the moving average version of the learned variables for eval.
saver = get_saver(MOVING_AVERAGE_DECAY, "bn_")
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
graph_def = tf.get_default_graph().as_graph_def()
summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir,
graph_def=graph_def)
while True:
eval_once(saver, summary_writer, top_k_ops, summary_op)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
