def get_data_from_tfrecords(dataset_name, dataset_split, tfrecords_dir,
batch_size, epoch):
dataset = segmentation_dataset.get_dataset(dataset_name, dataset_split,
tfrecords_dir)
if dataset_split == 'train':
is_training = True
else:
is_training = False
data_provider = dataset_data_provider.DatasetDataProvider(
dataset,
num_readers=_NUM_READERS,
num_epochs=epoch if is_training else 1,
shuffle=is_training)
[image, image_mask,
image_name] = data_provider.get(['image', 'labels_class', 'image_name'])
image.set_shape([1024, 1024, 3])
image_mask.set_shape([1024, 1024, 1])
input_images, input_image_masks, input_image_image_names = tf.train.batch(
[image, image_mask, image_name],
batch_size=batch_size,
num_threads=_NUM_PREPROCESSING_THREADS,
capacity=200)
batch_queue = slim.prefetch_queue.prefetch_queue(
[input_images, input_image_masks, input_image_image_names],
capacity=20)
input_images, input_image_masks, input_image_image_names = batch_queue.dequeue(
)
input_images = normalize_images(input_images)
return input_images, input_image_masks, input_image_image_names
def main(unused_argv):
tf.logging.set_verbosity(tf.logging.INFO)
# Set up deployment (i.e., multi-GPUs and/or multi-replicas).
config = model_deploy.DeploymentConfig(num_clones=FLAGS.num_clones,
clone_on_cpu=FLAGS.clone_on_cpu,
replica_id=FLAGS.task,
num_replicas=FLAGS.num_replicas,
num_ps_tasks=FLAGS.num_ps_tasks)
# Split the batch across GPUs.
assert FLAGS.train_batch_size % config.num_clones == 0, (
'Training batch size not divisble by number of clones (GPUs).')
clone_batch_size = FLAGS.train_batch_size // config.num_clones
# Get dataset-dependent information.
dataset = segmentation_dataset.get_dataset(FLAGS.dataset,
FLAGS.train_split,
dataset_dir=FLAGS.dataset_dir)
tf.gfile.MakeDirs(FLAGS.train_logdir)
tf.logging.info('Training on %s set', FLAGS.train_split)
with tf.Graph().as_default() as graph:
with tf.device(config.inputs_device()):
samples = input_generator.get(
dataset,
FLAGS.train_crop_size,
clone_batch_size,
min_resize_value=FLAGS.min_resize_value,
max_resize_value=FLAGS.max_resize_value,
resize_factor=FLAGS.resize_factor,
min_scale_factor=FLAGS.min_scale_factor,
max_scale_factor=FLAGS.max_scale_factor,
scale_factor_step_size=FLAGS.scale_factor_step_size,
dataset_split=FLAGS.train_split,
is_training=True,
model_variant=FLAGS.model_variant,
num_readers=8,
num_threads=8)
inputs_queue = prefetch_queue.prefetch_queue(samples,
capacity=128 *
config.num_clones)
# Create the global step on the device storing the variables.
with tf.device(config.variables_device()):
global_step = tf.train.get_or_create_global_step()
# Define the model and create clones.
model_fn = _build_network
model_args = (inputs_queue, {
common.OUTPUT_TYPE: dataset.num_classes
}, dataset.ignore_label)
clones = model_deploy.create_clones(config,
model_fn,
args=model_args)
# Gather update_ops from the first clone. These contain, for example,
# the updates for the batch_norm variables created by model_fn.
first_clone_scope = config.clone_scope(0)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS,
first_clone_scope)
# Gather initial summaries.
summaries = set(tf.get_collection(tf.GraphKeys.SUMMARIES))
# Add summaries for model variables.
for model_var in slim.get_model_variables():
summaries.add(tf.summary.histogram(model_var.op.name, model_var))
# Add summaries for images, labels, semantic predictions
if FLAGS.save_summaries_images:
summary_image = graph.get_tensor_by_name(
('%s/%s:0' % (first_clone_scope, common.IMAGE)).strip('/'))
summaries.add(
tf.summary.image('samples/%s' % common.IMAGE, summary_image))
first_clone_label = graph.get_tensor_by_name(
('%s/%s:0' % (first_clone_scope, common.LABEL)).strip('/'))
# Scale up summary image pixel values for better visualization.
pixel_scaling = max(1, 255 // dataset.num_classes)
summary_label = tf.cast(first_clone_label * pixel_scaling,
tf.uint8)
summaries.add(
tf.summary.image('samples/%s' % common.LABEL, summary_label))
first_clone_output = graph.get_tensor_by_name(
('%s/%s:0' %
(first_clone_scope, common.OUTPUT_TYPE)).strip('/'))
predictions = tf.expand_dims(tf.argmax(first_clone_output, 3), -1)
summary_predictions = tf.cast(predictions * pixel_scaling,
tf.uint8)
summaries.add(
tf.summary.image('samples/%s' % common.OUTPUT_TYPE,
summary_predictions))
# Add summaries for losses.
for loss in tf.get_collection(tf.GraphKeys.LOSSES, first_clone_scope):
summaries.add(tf.summary.scalar('losses/%s' % loss.op.name, loss))
# Build the optimizer based on the device specification.
with tf.device(config.optimizer_device()):
learning_rate = train_utils.get_model_learning_rate(
FLAGS.learning_policy, FLAGS.base_learning_rate,
FLAGS.learning_rate_decay_step,
FLAGS.learning_rate_decay_factor,
FLAGS.training_number_of_steps, FLAGS.learning_power,
FLAGS.slow_start_step, FLAGS.slow_start_learning_rate)
optimizer = tf.train.AdamOptimizer(learning_rate)
summaries.add(tf.summary.scalar('learning_rate', learning_rate))
startup_delay_steps = FLAGS.task * FLAGS.startup_delay_steps
for variable in slim.get_model_variables():
summaries.add(tf.summary.histogram(variable.op.name, variable))
with tf.device(config.variables_device()):
total_loss, grads_and_vars = model_deploy.optimize_clones(
clones, optimizer)
total_loss = tf.check_numerics(total_loss, 'Loss is inf or nan.')
summaries.add(tf.summary.scalar('total_loss', total_loss))
# Modify the gradients for biases and last layer variables.
last_layers = model.get_extra_layer_scopes(
FLAGS.last_layers_contain_logits_only)
grad_mult = train_utils.get_model_gradient_multipliers(
last_layers, FLAGS.last_layer_gradient_multiplier)
if grad_mult:
grads_and_vars = slim.learning.multiply_gradients(
grads_and_vars, grad_mult)
# Create gradient update op.
grad_updates = optimizer.apply_gradients(grads_and_vars,
global_step=global_step)
update_ops.append(grad_updates)
update_op = tf.group(*update_ops)
with tf.control_dependencies([update_op]):
train_tensor = tf.identity(total_loss, name='train_op')
# Add the summaries from the first clone. These contain the summaries
# created by model_fn and either optimize_clones() or _gather_clone_loss().
summaries |= set(
tf.get_collection(tf.GraphKeys.SUMMARIES, first_clone_scope))
# Merge all summaries together.
summary_op = tf.summary.merge(list(summaries))
# Soft placement allows placing on CPU ops without GPU implementation.
session_config = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)
if FLAGS.per_process_gpu_memory_fraction is not None:
session_config.gpu_options.per_process_gpu_memory_fraction = FLAGS.per_process_gpu_memory_fraction
# Start the training.
slim.learning.train(train_tensor,
logdir=FLAGS.train_logdir,
log_every_n_steps=FLAGS.log_steps,
master=FLAGS.master,
number_of_steps=FLAGS.training_number_of_steps,
is_chief=(FLAGS.task == 0),
session_config=session_config,
startup_delay_steps=startup_delay_steps,
init_fn=train_utils.get_model_init_fn(
FLAGS.train_logdir,
FLAGS.tf_initial_checkpoint,
FLAGS.initialize_last_layer,
last_layers,
ignore_missing_vars=True),
summary_op=summary_op,
save_summaries_secs=FLAGS.save_summaries_secs,
save_interval_secs=FLAGS.save_interval_secs)
def main(unused_argv):
tf.logging.set_verbosity(tf.logging.INFO)
# Get dataset-dependent information.
dataset = segmentation_dataset.get_dataset(FLAGS.dataset,
FLAGS.vis_split,
dataset_dir=FLAGS.dataset_dir)
train_id_to_eval_id = None
if dataset.name == segmentation_dataset.get_cityscapes_dataset_name():
tf.logging.info('Cityscapes requires converting train_id to eval_id.')
train_id_to_eval_id = _CITYSCAPES_TRAIN_ID_TO_EVAL_ID
# Prepare for visualization.
tf.gfile.MakeDirs(FLAGS.vis_logdir)
save_dir = os.path.join(FLAGS.vis_logdir, _SEMANTIC_PREDICTION_SAVE_FOLDER)
tf.gfile.MakeDirs(save_dir)
raw_save_dir = os.path.join(FLAGS.vis_logdir,
_RAW_SEMANTIC_PREDICTION_SAVE_FOLDER)
tf.gfile.MakeDirs(raw_save_dir)
tf.logging.info('Visualizing on %s set', FLAGS.vis_split)
g = tf.Graph()
with g.as_default():
samples = input_generator.get(dataset,
FLAGS.vis_crop_size,
FLAGS.vis_batch_size,
min_resize_value=FLAGS.min_resize_value,
max_resize_value=FLAGS.max_resize_value,
resize_factor=FLAGS.resize_factor,
dataset_split=FLAGS.vis_split,
is_training=False,
model_variant=FLAGS.model_variant)
model_options = common.ModelOptions(
outputs_to_num_classes={common.OUTPUT_TYPE: dataset.num_classes},
crop_size=FLAGS.vis_crop_size,
atrous_rates=FLAGS.atrous_rates,
output_stride=FLAGS.output_stride)
if tuple(FLAGS.eval_scales) == (1.0, ):
tf.logging.info('Performing single-scale test.')
predictions = model.predict_labels(
samples[common.IMAGE],
model_options=model_options,
image_pyramid=FLAGS.image_pyramid)
else:
tf.logging.info('Performing multi-scale test.')
predictions = model.predict_labels_multi_scale(
samples[common.IMAGE],
model_options=model_options,
eval_scales=FLAGS.eval_scales,
add_flipped_images=FLAGS.add_flipped_images)
predictions = predictions[common.OUTPUT_TYPE]
if FLAGS.min_resize_value and FLAGS.max_resize_value:
# Only support batch_size = 1, since we assume the dimensions of original
# image after tf.squeeze is [height, width, 3].
assert FLAGS.vis_batch_size == 1
# Reverse the resizing and padding operations performed in preprocessing.
# First, we slice the valid regions (i.e., remove padded region) and then
# we reisze the predictions back.
original_image = tf.squeeze(samples[common.ORIGINAL_IMAGE])
original_image_shape = tf.shape(original_image)
predictions = tf.slice(
predictions, [0, 0, 0],
[1, original_image_shape[0], original_image_shape[1]])
resized_shape = tf.cast([
tf.squeeze(samples[common.HEIGHT]),
tf.squeeze(samples[common.WIDTH])
], tf.int32)
predictions = tf.squeeze(
tf.image.resize_images(
tf.expand_dims(predictions, 3),
resized_shape,
method=tf.image.ResizeMethod.NEAREST_NEIGHBOR,
align_corners=True), 3)
labels = samples[common.LABEL]
labels = tf.where(tf.equal(labels, dataset.ignore_label),
tf.ones_like(labels) * 20, labels)
tf.train.get_or_create_global_step()
saver = tf.train.Saver(slim.get_variables_to_restore())
sv = tf.train.Supervisor(graph=g,
logdir=FLAGS.vis_logdir,
init_op=tf.global_variables_initializer(),
summary_op=None,
summary_writer=None,
global_step=None,
saver=saver)
num_batches = int(
math.ceil(dataset.num_samples / float(FLAGS.vis_batch_size)))
last_checkpoint = None
# Loop to visualize the results when new checkpoint is created.
num_iters = 0
while (FLAGS.max_number_of_iterations <= 0
or num_iters < FLAGS.max_number_of_iterations):
num_iters += 1
last_checkpoint = slim.evaluation.wait_for_new_checkpoint(
FLAGS.checkpoint_dir, last_checkpoint)
start = time.time()
tf.logging.info('Starting visualization at ' +
time.strftime('%Y-%m-%d-%H:%M:%S', time.gmtime()))
tf.logging.info('Visualizing with model %s', last_checkpoint)
with sv.managed_session(FLAGS.master,
start_standard_services=False) as sess:
sv.start_queue_runners(sess)
sv.saver.restore(sess, last_checkpoint)
image_id_offset = 0
for batch in range(num_batches):
tf.logging.info('Visualizing batch %d / %d', batch + 1,
num_batches)
_process_batch(
sess=sess,
original_images=samples[common.ORIGINAL_IMAGE],
semantic_predictions=predictions,
image_names=samples[common.IMAGE_NAME],
image_heights=samples[common.HEIGHT],
image_widths=samples[common.WIDTH],
labels=labels,
image_id_offset=image_id_offset,
save_dir=save_dir,
raw_save_dir=raw_save_dir,
train_id_to_eval_id=train_id_to_eval_id)
image_id_offset += FLAGS.vis_batch_size
tf.logging.info('Finished visualization at ' +
time.strftime('%Y-%m-%d-%H:%M:%S', time.gmtime()))
time_to_next_eval = start + FLAGS.eval_interval_secs - time.time()
if time_to_next_eval > 0:
time.sleep(time_to_next_eval)
def main(unused_argv):
tf.logging.set_verbosity(tf.logging.INFO)
# Get dataset-dependent information.
dataset = segmentation_dataset.get_dataset(FLAGS.dataset,
FLAGS.eval_split,
dataset_dir=FLAGS.dataset_dir)
tf.gfile.MakeDirs(FLAGS.eval_logdir)
tf.logging.info('Evaluating on %s set', FLAGS.eval_split)
with tf.Graph().as_default():
samples = input_generator.get(dataset,
FLAGS.eval_crop_size,
FLAGS.eval_batch_size,
min_resize_value=FLAGS.min_resize_value,
max_resize_value=FLAGS.max_resize_value,
resize_factor=FLAGS.resize_factor,
dataset_split=FLAGS.eval_split,
is_training=False,
model_variant=FLAGS.model_variant)
model_options = common.ModelOptions(
outputs_to_num_classes={common.OUTPUT_TYPE: dataset.num_classes},
crop_size=FLAGS.eval_crop_size,
atrous_rates=FLAGS.atrous_rates,
output_stride=FLAGS.output_stride)
if tuple(FLAGS.eval_scales) == (1.0, ):
tf.logging.info('Performing single-scale test.')
predictions = model.predict_labels(
samples[common.IMAGE],
model_options,
image_pyramid=FLAGS.image_pyramid)
else:
tf.logging.info('Performing multi-scale test.')
predictions = model.predict_labels_multi_scale(
samples[common.IMAGE],
model_options=model_options,
eval_scales=FLAGS.eval_scales,
add_flipped_images=FLAGS.add_flipped_images)
predictions = predictions[common.OUTPUT_TYPE]
predictions = tf.reshape(predictions, shape=[-1])
labels = tf.reshape(samples[common.LABEL], shape=[-1])
weights = tf.cast(tf.not_equal(labels, dataset.ignore_label),
tf.float32)
# Set ignore_label regions to label 0, because metrics.mean_iou requires
# range of labels = [0, dataset.num_classes). Note the ignore_label regions
# are not evaluated since the corresponding regions contain weights = 0.
labels = tf.where(tf.equal(labels, dataset.ignore_label),
tf.zeros_like(labels), labels)
predictions_tag = 'miou'
for eval_scale in FLAGS.eval_scales:
predictions_tag += '_' + str(eval_scale)
if FLAGS.add_flipped_images:
predictions_tag += '_flipped'
# Define the evaluation metric.
metric_map = {}
metric_map[predictions_tag] = tf.metrics.mean_iou(predictions,
labels,
dataset.num_classes,
weights=weights)
metrics_to_values, metrics_to_updates = (
tf.contrib.metrics.aggregate_metric_map(metric_map))
for metric_name, metric_value in six.iteritems(metrics_to_values):
slim.summaries.add_scalar_summary(metric_value,
metric_name,
print_summary=True)
num_batches = int(
math.ceil(dataset.num_samples / float(FLAGS.eval_batch_size)))
tf.logging.info('Eval num images %d', dataset.num_samples)
tf.logging.info('Eval batch size %d and num batch %d',
FLAGS.eval_batch_size, num_batches)
num_eval_iters = None
if FLAGS.max_number_of_evaluations > 0:
num_eval_iters = FLAGS.max_number_of_evaluations
# Soft placement allows placing on CPU ops without GPU implementation.
session_config = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)
session_config.gpu_options.allow_growth = True
slim.evaluation.evaluation_loop(
session_config=session_config,
master=FLAGS.master,
checkpoint_dir=FLAGS.checkpoint_dir,
logdir=FLAGS.eval_logdir,
num_evals=num_batches,
eval_op=list(metrics_to_updates.values()),
max_number_of_evaluations=num_eval_iters,
eval_interval_secs=FLAGS.eval_interval_secs)