Skip to content

GitLab

Commit f906646c authored by Duc Nguyen's avatar Duc Nguyen Committed by GitHub
Browse files

Merge branch 'master' into patch-2

parents 2f3666ed a6df5573
Hide whitespace changes
Inline Side-by-side
Showing with 448 additions and 149 deletions
object_detection/core/data_decoder.py
View file @ f906646c
......@@ -26,9 +26,8 @@ class DataDecoder(object):
"""Interface for data decoders."""
__metaclass__ = ABCMeta
# TODO: snake_case this method.
@abstractmethod
def Decode(self, data):
def decode(self, data):
"""Return a single image and associated labels.
Args:
......
object_detection/core/post_processing.py
View file @ f906646c
......@@ -131,7 +131,7 @@ def multiclass_non_max_suppression(boxes,
boxlist_and_class_scores.add_field(fields.BoxListFields.masks,
per_class_masks)
if additional_fields is not None:
for key, tensor in additional_fields.iteritems():
for key, tensor in additional_fields.items():
boxlist_and_class_scores.add_field(key, tensor)
boxlist_filtered = box_list_ops.filter_greater_than(
boxlist_and_class_scores, score_thresh)
......
object_detection/core/prefetcher.py
View file @ f906646c
......@@ -45,7 +45,7 @@ def prefetch(tensor_dict, capacity):
Returns:
a FIFO prefetcher queue
"""
names = tensor_dict.keys()
names = list(tensor_dict.keys())
dtypes = [t.dtype for t in tensor_dict.values()]
shapes = [t.get_shape() for t in tensor_dict.values()]
prefetch_queue = tf.PaddingFIFOQueue(capacity, dtypes=dtypes,
......
object_detection/core/preprocessor.py
View file @ f906646c
......@@ -341,7 +341,6 @@ def random_pixel_value_scale(image, minval=0.9, maxval=1.1, seed=None):
Returns:
image: image which is the same shape as input image.
boxes: boxes which is the same shape as input boxes.
"""
with tf.name_scope('RandomPixelValueScale', values=[image]):
color_coef = tf.random_uniform(
......
object_detection/core/preprocessor_test.py
View file @ f906646c
......@@ -15,14 +15,19 @@
"""Tests for object_detection.core.preprocessor."""
import mock
import numpy as np
import six
import tensorflow as tf
from object_detection.core import preprocessor
from object_detection.core import standard_fields as fields
if six.PY2:
import mock # pylint: disable=g-import-not-at-top
else:
from unittest import mock # pylint: disable=g-import-not-at-top
class PreprocessorTest(tf.test.TestCase):
......
object_detection/create_pascal_tf_record.py
View file @ f906646c
......@@ -83,7 +83,7 @@ def dict_to_tf_example(data,
"""
img_path = os.path.join(data['folder'], image_subdirectory, data['filename'])
full_path = os.path.join(dataset_directory, img_path)
with tf.gfile.GFile(full_path) as fid:
with tf.gfile.GFile(full_path, 'rb') as fid:
encoded_jpg = fid.read()
encoded_jpg_io = io.BytesIO(encoded_jpg)
image = PIL.Image.open(encoded_jpg_io)
......@@ -114,19 +114,21 @@ def dict_to_tf_example(data,
ymin.append(float(obj['bndbox']['ymin']) / height)
xmax.append(float(obj['bndbox']['xmax']) / width)
ymax.append(float(obj['bndbox']['ymax']) / height)
classes_text.append(obj['name'])
classes_text.append(obj['name'].encode('utf8'))
classes.append(label_map_dict[obj['name']])
truncated.append(int(obj['truncated']))
poses.append(obj['pose'])
poses.append(obj['pose'].encode('utf8'))
example = tf.train.Example(features=tf.train.Features(feature={
'image/height': dataset_util.int64_feature(height),
'image/width': dataset_util.int64_feature(width),
'image/filename': dataset_util.bytes_feature(data['filename']),
'image/source_id': dataset_util.bytes_feature(data['filename']),
'image/key/sha256': dataset_util.bytes_feature(key),
'image/filename': dataset_util.bytes_feature(
data['filename'].encode('utf8')),
'image/source_id': dataset_util.bytes_feature(
data['filename'].encode('utf8')),
'image/key/sha256': dataset_util.bytes_feature(key.encode('utf8')),
'image/encoded': dataset_util.bytes_feature(encoded_jpg),
'image/format': dataset_util.bytes_feature('jpeg'),
'image/format': dataset_util.bytes_feature('jpeg'.encode('utf8')),
'image/object/bbox/xmin': dataset_util.float_list_feature(xmin),
'image/object/bbox/xmax': dataset_util.float_list_feature(xmax),
'image/object/bbox/ymin': dataset_util.float_list_feature(ymin),
......
object_detection/create_pet_tf_record.py
View file @ f906646c
......@@ -86,7 +86,7 @@ def dict_to_tf_example(data,
ValueError: if the image pointed to by data['filename'] is not a valid JPEG
"""
img_path = os.path.join(image_subdirectory, data['filename'])
with tf.gfile.GFile(img_path) as fid:
with tf.gfile.GFile(img_path, 'rb') as fid:
encoded_jpg = fid.read()
encoded_jpg_io = io.BytesIO(encoded_jpg)
image = PIL.Image.open(encoded_jpg_io)
......@@ -118,19 +118,21 @@ def dict_to_tf_example(data,
xmax.append(float(obj['bndbox']['xmax']) / width)
ymax.append(float(obj['bndbox']['ymax']) / height)
class_name = get_class_name_from_filename(data['filename'])
classes_text.append(class_name)
classes_text.append(class_name.encode('utf8'))
classes.append(label_map_dict[class_name])
truncated.append(int(obj['truncated']))
poses.append(obj['pose'])
poses.append(obj['pose'].encode('utf8'))
example = tf.train.Example(features=tf.train.Features(feature={
'image/height': dataset_util.int64_feature(height),
'image/width': dataset_util.int64_feature(width),
'image/filename': dataset_util.bytes_feature(data['filename']),
'image/source_id': dataset_util.bytes_feature(data['filename']),
'image/key/sha256': dataset_util.bytes_feature(key),
'image/filename': dataset_util.bytes_feature(
data['filename'].encode('utf8')),
'image/source_id': dataset_util.bytes_feature(
data['filename'].encode('utf8')),
'image/key/sha256': dataset_util.bytes_feature(key.encode('utf8')),
'image/encoded': dataset_util.bytes_feature(encoded_jpg),
'image/format': dataset_util.bytes_feature('jpeg'),
'image/format': dataset_util.bytes_feature('jpeg'.encode('utf8')),
'image/object/bbox/xmin': dataset_util.float_list_feature(xmin),
'image/object/bbox/xmax': dataset_util.float_list_feature(xmax),
'image/object/bbox/ymin': dataset_util.float_list_feature(ymin),
......
object_detection/data_decoders/tf_example_decoder.py
View file @ f906646c
......@@ -82,7 +82,7 @@ class TfExampleDecoder(data_decoder.DataDecoder):
slim_example_decoder.Tensor('image/segmentation/object/class')),
}
def Decode(self, tf_example_string_tensor):
def decode(self, tf_example_string_tensor):
"""Decodes serialized tensorflow example and returns a tensor dictionary.
Args:
......
object_detection/data_decoders/tf_example_decoder_test.py
View file @ f906646c
......@@ -64,7 +64,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder()
tensor_dict = example_decoder.Decode(tf.convert_to_tensor(example))
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
self.assertAllEqual((tensor_dict[fields.InputDataFields.image].
get_shape().as_list()), [None, None, 3])
......@@ -84,7 +84,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder()
tensor_dict = example_decoder.Decode(tf.convert_to_tensor(example))
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
with self.test_session() as sess:
tensor_dict = sess.run(tensor_dict)
......@@ -103,7 +103,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder()
tensor_dict = example_decoder.Decode(tf.convert_to_tensor(example))
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
self.assertAllEqual((tensor_dict[fields.InputDataFields.image].
get_shape().as_list()), [None, None, 3])
......@@ -130,7 +130,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder()
tensor_dict = example_decoder.Decode(tf.convert_to_tensor(example))
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
self.assertAllEqual((tensor_dict[fields.InputDataFields.groundtruth_boxes].
get_shape().as_list()), [None, 4])
......@@ -153,7 +153,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder()
tensor_dict = example_decoder.Decode(tf.convert_to_tensor(example))
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
self.assertAllEqual((tensor_dict[
fields.InputDataFields.groundtruth_classes].get_shape().as_list()),
......@@ -176,7 +176,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder()
tensor_dict = example_decoder.Decode(tf.convert_to_tensor(example))
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
self.assertAllEqual((tensor_dict[fields.InputDataFields.groundtruth_area].
get_shape().as_list()), [None])
......@@ -197,7 +197,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder()
tensor_dict = example_decoder.Decode(tf.convert_to_tensor(example))
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
self.assertAllEqual((tensor_dict[
fields.InputDataFields.groundtruth_is_crowd].get_shape().as_list()),
......@@ -220,7 +220,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder()
tensor_dict = example_decoder.Decode(tf.convert_to_tensor(example))
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
self.assertAllEqual((tensor_dict[
fields.InputDataFields.groundtruth_difficult].get_shape().as_list()),
......@@ -263,7 +263,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
'image/segmentation/object/class': self._Int64Feature(
instance_segmentation_classes)})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder()
tensor_dict = example_decoder.Decode(tf.convert_to_tensor(example))
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
self.assertAllEqual((
tensor_dict[fields.InputDataFields.groundtruth_instance_masks].
......
object_detection/evaluator.py
View file @ f906646c
......@@ -154,7 +154,7 @@ def evaluate(create_input_dict_fn, create_model_fn, eval_config, categories,
"""
if batch_index >= eval_config.num_visualizations:
if 'original_image' in tensor_dict:
tensor_dict = {k: v for (k, v) in tensor_dict.iteritems()
tensor_dict = {k: v for (k, v) in tensor_dict.items()
if k != 'original_image'}
try:
(result_dict, _) = sess.run([tensor_dict, update_op])
......
object_detection/export_inference_graph.py
View file @ f906646c
......@@ -16,23 +16,30 @@
r"""Tool to export an object detection model for inference.
Prepares an object detection tensorflow graph for inference using model
configuration and an optional trained checkpoint.
configuration and an optional trained checkpoint. Outputs either an inference
graph or a SavedModel (https://tensorflow.github.io/serving/serving_basic.html).
The inference graph contains one of two input nodes depending on the user
The inference graph contains one of three input nodes depending on the user
specified option.
* `image_tensor`: Accepts a uint8 4-D tensor of shape [1, None, None, 3]
* `encoded_image_string_tensor`: Accepts a scalar string tensor of encoded PNG
or JPEG image.
* `tf_example`: Accepts a serialized TFExample proto. The batch size in this
case is always 1.
and the following output nodes:
* `num_detections` : Outputs float32 tensors of the form [batch]
and the following output nodes returned by the model.postprocess(..):
* `num_detections`: Outputs float32 tensors of the form [batch]
that specifies the number of valid boxes per image in the batch.
* `detection_boxes` : Outputs float32 tensors of the form
* `detection_boxes`: Outputs float32 tensors of the form
[batch, num_boxes, 4] containing detected boxes.
* `detection_scores` : Outputs float32 tensors of the form
* `detection_scores`: Outputs float32 tensors of the form
[batch, num_boxes] containing class scores for the detections.
* `detection_classes`: Outputs float32 tensors of the form
[batch, num_boxes] containing classes for the detections.
* `detection_masks`: Outputs float32 tensors of the form
[batch, num_boxes, mask_height, mask_width] containing predicted instance
masks for each box if its present in the dictionary of postprocessed
tensors returned by the model.
Note that currently `batch` is always 1, but we will support `batch` > 1 in
the future.
......@@ -61,7 +68,8 @@ slim = tf.contrib.slim
flags = tf.app.flags
flags.DEFINE_string('input_type', 'image_tensor', 'Type of input node. Can be '
'one of [`image_tensor` `tf_example_proto`]')
'one of [`image_tensor`, `encoded_image_string_tensor`, '
'`tf_example`]')
flags.DEFINE_string('pipeline_config_path', '',
'Path to a pipeline_pb2.TrainEvalPipelineConfig config '
'file.')
......@@ -70,6 +78,8 @@ flags.DEFINE_string('checkpoint_path', '', 'Optional path to checkpoint file. '
'the graph.')
flags.DEFINE_string('inference_graph_path', '', 'Path to write the output '
'inference graph.')
flags.DEFINE_bool('export_as_saved_model', False, 'Whether the exported graph '
'should be saved as a SavedModel')
FLAGS = flags.FLAGS
......@@ -83,7 +93,8 @@ def main(_):
text_format.Merge(f.read(), pipeline_config)
exporter.export_inference_graph(FLAGS.input_type, pipeline_config,
FLAGS.checkpoint_path,
FLAGS.inference_graph_path)
FLAGS.inference_graph_path,
FLAGS.export_as_saved_model)
if __name__ == '__main__':
......
object_detection/exporter.py
View file @ f906646c
......@@ -22,6 +22,7 @@ from tensorflow.python.client import session
from tensorflow.python.framework import graph_util
from tensorflow.python.framework import importer
from tensorflow.python.platform import gfile
from tensorflow.python.saved_model import signature_constants
from tensorflow.python.training import saver as saver_lib
from object_detection.builders import model_builder
from object_detection.core import standard_fields as fields
......@@ -30,8 +31,8 @@ from object_detection.data_decoders import tf_example_decoder
slim = tf.contrib.slim
# TODO: Replace with freeze_graph.freeze_graph_with_def_protos when newer
# version of Tensorflow becomes more common.
# TODO: Replace with freeze_graph.freeze_graph_with_def_protos when
# newer version of Tensorflow becomes more common.
def freeze_graph_with_def_protos(
input_graph_def,
input_saver_def,
......@@ -39,7 +40,6 @@ def freeze_graph_with_def_protos(
output_node_names,
restore_op_name,
filename_tensor_name,
output_graph,
clear_devices,
initializer_nodes,
variable_names_blacklist=''):
......@@ -48,12 +48,12 @@ def freeze_graph_with_def_protos(
# 'input_checkpoint' may be a prefix if we're using Saver V2 format
if not saver_lib.checkpoint_exists(input_checkpoint):
logging.info('Input checkpoint "' + input_checkpoint + '" does not exist!')
return -1
raise ValueError(
'Input checkpoint "' + input_checkpoint + '" does not exist!')
if not output_node_names:
logging.info('You must supply the name of a node to --output_node_names.')
return -1
raise ValueError(
'You must supply the name of a node to --output_node_names.')
# Remove all the explicit device specifications for this node. This helps to
# make the graph more portable.
......@@ -92,16 +92,37 @@ def freeze_graph_with_def_protos(
output_node_names.split(','),
variable_names_blacklist=variable_names_blacklist)
with gfile.GFile(output_graph, 'wb') as f:
f.write(output_graph_def.SerializeToString())
logging.info('%d ops in the final graph.', len(output_graph_def.node))
return output_graph_def
def get_frozen_graph_def(inference_graph_def, use_moving_averages,
input_checkpoint, output_node_names):
"""Freezes all variables in a graph definition."""
saver = None
if use_moving_averages:
variable_averages = tf.train.ExponentialMovingAverage(0.0)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
else:
saver = tf.train.Saver()
frozen_graph_def = freeze_graph_with_def_protos(
input_graph_def=inference_graph_def,
input_saver_def=saver.as_saver_def(),
input_checkpoint=input_checkpoint,
output_node_names=output_node_names,
restore_op_name='save/restore_all',
filename_tensor_name='save/Const:0',
clear_devices=True,
initializer_nodes='')
return frozen_graph_def
# TODO: Support batch tf example inputs.
def _tf_example_input_placeholder():
tf_example_placeholder = tf.placeholder(
tf.string, shape=[], name='tf_example')
tensor_dict = tf_example_decoder.TfExampleDecoder().Decode(
tensor_dict = tf_example_decoder.TfExampleDecoder().decode(
tf_example_placeholder)
image = tensor_dict[fields.InputDataFields.image]
return tf.expand_dims(image, axis=0)
......@@ -112,9 +133,21 @@ def _image_tensor_input_placeholder():
shape=(1, None, None, 3),
name='image_tensor')
def _encoded_image_string_tensor_input_placeholder():
image_str = tf.placeholder(dtype=tf.string,
shape=[],
name='encoded_image_string_tensor')
image_tensor = tf.image.decode_image(image_str, channels=3)
image_tensor.set_shape((None, None, 3))
return tf.expand_dims(image_tensor, axis=0)
input_placeholder_fn_map = {
'image_tensor': _image_tensor_input_placeholder,
'encoded_image_string_tensor':
_encoded_image_string_tensor_input_placeholder,
'tf_example': _tf_example_input_placeholder,
'image_tensor': _image_tensor_input_placeholder
}
......@@ -129,23 +162,36 @@ def _add_output_tensor_nodes(postprocessed_tensors):
containing scores for the detected boxes.
* detection_classes: float32 tensor of shape [batch_size, num_boxes]
containing class predictions for the detected boxes.
* detection_masks: (Optional) float32 tensor of shape
[batch_size, num_boxes, mask_height, mask_width] containing masks for each
detection box.
Args:
postprocessed_tensors: a dictionary containing the following fields
'detection_boxes': [batch, max_detections, 4]
'detection_scores': [batch, max_detections]
'detection_classes': [batch, max_detections]
'detection_masks': [batch, max_detections, mask_height, mask_width]
(optional).
'num_detections': [batch]
Returns:
A tensor dict containing the added output tensor nodes.
"""
label_id_offset = 1
boxes = postprocessed_tensors.get('detection_boxes')
scores = postprocessed_tensors.get('detection_scores')
classes = postprocessed_tensors.get('detection_classes') + label_id_offset
masks = postprocessed_tensors.get('detection_masks')
num_detections = postprocessed_tensors.get('num_detections')
tf.identity(boxes, name='detection_boxes')
tf.identity(scores, name='detection_scores')
tf.identity(classes, name='detection_classes')
tf.identity(num_detections, name='num_detections')
outputs = {}
outputs['detection_boxes'] = tf.identity(boxes, name='detection_boxes')
outputs['detection_scores'] = tf.identity(scores, name='detection_scores')
outputs['detection_classes'] = tf.identity(classes, name='detection_classes')
outputs['num_detections'] = tf.identity(num_detections, name='num_detections')
if masks is not None:
outputs['detection_masks'] = tf.identity(masks, name='detection_masks')
return outputs
def _write_inference_graph(inference_graph_path,
......@@ -172,23 +218,17 @@ def _write_inference_graph(inference_graph_path,
"""
inference_graph_def = tf.get_default_graph().as_graph_def()
if checkpoint_path:
saver = None
if use_moving_averages:
variable_averages = tf.train.ExponentialMovingAverage(0.0)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
else:
saver = tf.train.Saver()
freeze_graph_with_def_protos(
input_graph_def=inference_graph_def,
input_saver_def=saver.as_saver_def(),
output_graph_def = get_frozen_graph_def(
inference_graph_def=inference_graph_def,
use_moving_averages=use_moving_averages,
input_checkpoint=checkpoint_path,
output_node_names=output_node_names,
restore_op_name='save/restore_all',
filename_tensor_name='save/Const:0',
output_graph=inference_graph_path,
clear_devices=True,
initializer_nodes='')
)
with gfile.GFile(inference_graph_path, 'wb') as f:
f.write(output_graph_def.SerializeToString())
logging.info('%d ops in the final graph.', len(output_graph_def.node))
return
tf.train.write_graph(inference_graph_def,
os.path.dirname(inference_graph_path),
......@@ -196,24 +236,90 @@ def _write_inference_graph(inference_graph_path,
as_text=False)
def _write_saved_model(inference_graph_path, inputs, outputs,
checkpoint_path=None, use_moving_averages=False):
"""Writes SavedModel to disk.
If checkpoint_path is not None bakes the weights into the graph thereby
eliminating the need of checkpoint files during inference. If the model
was trained with moving averages, setting use_moving_averages to true
restores the moving averages, otherwise the original set of variables
is restored.
Args:
inference_graph_path: Path to write inference graph.
inputs: The input image tensor to use for detection.
outputs: A tensor dictionary containing the outputs of a DetectionModel.
checkpoint_path: Optional path to the checkpoint file.
use_moving_averages: Whether to export the original or the moving averages
of the trainable variables from the checkpoint.
"""
inference_graph_def = tf.get_default_graph().as_graph_def()
checkpoint_graph_def = None
if checkpoint_path:
output_node_names = ','.join(outputs.keys())
checkpoint_graph_def = get_frozen_graph_def(
inference_graph_def=inference_graph_def,
use_moving_averages=use_moving_averages,
input_checkpoint=checkpoint_path,
output_node_names=output_node_names
)
with tf.Graph().as_default():
with session.Session() as sess:
tf.import_graph_def(checkpoint_graph_def)
builder = tf.saved_model.builder.SavedModelBuilder(inference_graph_path)
tensor_info_inputs = {
'inputs': tf.saved_model.utils.build_tensor_info(inputs)}
tensor_info_outputs = {}
for k, v in outputs.items():
tensor_info_outputs[k] = tf.saved_model.utils.build_tensor_info(v)
detection_signature = (
tf.saved_model.signature_def_utils.build_signature_def(
inputs=tensor_info_inputs,
outputs=tensor_info_outputs,
method_name=signature_constants.PREDICT_METHOD_NAME))
builder.add_meta_graph_and_variables(
sess, [tf.saved_model.tag_constants.SERVING],
signature_def_map={
'signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY':
detection_signature,
},
)
builder.save()
def _export_inference_graph(input_type,
detection_model,
use_moving_averages,
checkpoint_path,
inference_graph_path):
inference_graph_path,
export_as_saved_model=False):
"""Export helper."""
if input_type not in input_placeholder_fn_map:
raise ValueError('Unknown input type: {}'.format(input_type))
inputs = tf.to_float(input_placeholder_fn_map[input_type]())
preprocessed_inputs = detection_model.preprocess(inputs)
output_tensors = detection_model.predict(preprocessed_inputs)
postprocessed_tensors = detection_model.postprocess(output_tensors)
_add_output_tensor_nodes(postprocessed_tensors)
_write_inference_graph(inference_graph_path, checkpoint_path,
use_moving_averages)
outputs = _add_output_tensor_nodes(postprocessed_tensors)
out_node_names = list(outputs.keys())
if export_as_saved_model:
_write_saved_model(inference_graph_path, inputs, outputs, checkpoint_path,
use_moving_averages)
else:
_write_inference_graph(inference_graph_path, checkpoint_path,
use_moving_averages,
output_node_names=','.join(out_node_names))
def export_inference_graph(input_type, pipeline_config, checkpoint_path,
inference_graph_path):
inference_graph_path, export_as_saved_model=False):
"""Exports inference graph for the model specified in the pipeline config.
Args:
......@@ -222,9 +328,12 @@ def export_inference_graph(input_type, pipeline_config, checkpoint_path,
pipeline_config: pipeline_pb2.TrainAndEvalPipelineConfig proto.
checkpoint_path: Path to the checkpoint file to freeze.
inference_graph_path: Path to write inference graph to.
export_as_saved_model: If the model should be exported as a SavedModel. If
false, it is saved as an inference graph.
"""
detection_model = model_builder.build(pipeline_config.model,
is_training=False)
_export_inference_graph(input_type, detection_model,
pipeline_config.eval_config.use_moving_averages,
checkpoint_path, inference_graph_path)
checkpoint_path, inference_graph_path,
export_as_saved_model)
object_detection/exporter_test.py
View file @ f906646c
......@@ -15,35 +15,44 @@
"""Tests for object_detection.export_inference_graph."""
import os
import mock
import numpy as np
import six
import tensorflow as tf
from object_detection import exporter
from object_detection.builders import model_builder
from object_detection.core import model
from object_detection.protos import pipeline_pb2
if six.PY2:
import mock # pylint: disable=g-import-not-at-top
else:
from unittest import mock # pylint: disable=g-import-not-at-top
class FakeModel(model.DetectionModel):
def __init__(self, add_detection_masks=False):
self._add_detection_masks = add_detection_masks
def preprocess(self, inputs):
return (tf.identity(inputs) *
tf.get_variable('dummy', shape=(),
initializer=tf.constant_initializer(2),
dtype=tf.float32))
return tf.identity(inputs)
def predict(self, preprocessed_inputs):
return {'image': tf.identity(preprocessed_inputs)}
return {'image': tf.layers.conv2d(preprocessed_inputs, 3, 1)}
def postprocess(self, prediction_dict):
with tf.control_dependencies(prediction_dict.values()):
return {
postprocessed_tensors = {
'detection_boxes': tf.constant([[0.0, 0.0, 0.5, 0.5],
[0.5, 0.5, 0.8, 0.8]], tf.float32),
'detection_scores': tf.constant([[0.7, 0.6]], tf.float32),
'detection_classes': tf.constant([[0, 1]], tf.float32),
'num_detections': tf.constant([2], tf.float32)
}
if self._add_detection_masks:
postprocessed_tensors['detection_masks'] = tf.constant(
np.arange(32).reshape([2, 4, 4]), tf.float32)
return postprocessed_tensors
def restore_fn(self, checkpoint_path, from_detection_checkpoint):
pass
......@@ -58,8 +67,11 @@ class ExportInferenceGraphTest(tf.test.TestCase):
use_moving_averages):
g = tf.Graph()
with g.as_default():
mock_model = FakeModel(num_classes=1)
mock_model.preprocess(tf.constant([1, 3, 4, 3], tf.float32))
mock_model = FakeModel()
preprocessed_inputs = mock_model.preprocess(
tf.ones([1, 3, 4, 3], tf.float32))
predictions = mock_model.predict(preprocessed_inputs)
mock_model.postprocess(predictions)
if use_moving_averages:
tf.train.ExponentialMovingAverage(0.0).apply()
saver = tf.train.Saver()
......@@ -93,7 +105,7 @@ class ExportInferenceGraphTest(tf.test.TestCase):
def test_export_graph_with_image_tensor_input(self):
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(num_classes=1)
mock_builder.return_value = FakeModel()
inference_graph_path = os.path.join(self.get_temp_dir(),
'exported_graph.pbtxt')
......@@ -108,7 +120,7 @@ class ExportInferenceGraphTest(tf.test.TestCase):
def test_export_graph_with_tf_example_input(self):
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(num_classes=1)
mock_builder.return_value = FakeModel()
inference_graph_path = os.path.join(self.get_temp_dir(),
'exported_graph.pbtxt')
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
......@@ -119,6 +131,20 @@ class ExportInferenceGraphTest(tf.test.TestCase):
checkpoint_path=None,
inference_graph_path=inference_graph_path)
def test_export_graph_with_encoded_image_string_input(self):
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel()
inference_graph_path = os.path.join(self.get_temp_dir(),
'exported_graph.pbtxt')
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
pipeline_config.eval_config.use_moving_averages = False
exporter.export_inference_graph(
input_type='encoded_image_string_tensor',
pipeline_config=pipeline_config,
checkpoint_path=None,
inference_graph_path=inference_graph_path)
def test_export_frozen_graph(self):
checkpoint_path = os.path.join(self.get_temp_dir(), 'model-ckpt')
self._save_checkpoint_from_mock_model(checkpoint_path,
......@@ -127,7 +153,7 @@ class ExportInferenceGraphTest(tf.test.TestCase):
'exported_graph.pb')
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(num_classes=1)
mock_builder.return_value = FakeModel()
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
pipeline_config.eval_config.use_moving_averages = False
exporter.export_inference_graph(
......@@ -144,7 +170,7 @@ class ExportInferenceGraphTest(tf.test.TestCase):
'exported_graph.pb')
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(num_classes=1)
mock_builder.return_value = FakeModel()
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
pipeline_config.eval_config.use_moving_averages = True
exporter.export_inference_graph(
......@@ -153,6 +179,55 @@ class ExportInferenceGraphTest(tf.test.TestCase):
checkpoint_path=checkpoint_path,
inference_graph_path=inference_graph_path)
def test_export_model_with_all_output_nodes(self):
checkpoint_path = os.path.join(self.get_temp_dir(), 'model-ckpt')
self._save_checkpoint_from_mock_model(checkpoint_path,
use_moving_averages=False)
inference_graph_path = os.path.join(self.get_temp_dir(),
'exported_graph.pb')
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(add_detection_masks=True)
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
exporter.export_inference_graph(
input_type='image_tensor',
pipeline_config=pipeline_config,
checkpoint_path=checkpoint_path,
inference_graph_path=inference_graph_path)
inference_graph = self._load_inference_graph(inference_graph_path)
with self.test_session(graph=inference_graph):
inference_graph.get_tensor_by_name('image_tensor:0')
inference_graph.get_tensor_by_name('detection_boxes:0')
inference_graph.get_tensor_by_name('detection_scores:0')
inference_graph.get_tensor_by_name('detection_classes:0')
inference_graph.get_tensor_by_name('detection_masks:0')
inference_graph.get_tensor_by_name('num_detections:0')
def test_export_model_with_detection_only_nodes(self):
checkpoint_path = os.path.join(self.get_temp_dir(), 'model-ckpt')
self._save_checkpoint_from_mock_model(checkpoint_path,
use_moving_averages=False)
inference_graph_path = os.path.join(self.get_temp_dir(),
'exported_graph.pb')
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(add_detection_masks=False)
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
exporter.export_inference_graph(
input_type='image_tensor',
pipeline_config=pipeline_config,
checkpoint_path=checkpoint_path,
inference_graph_path=inference_graph_path)
inference_graph = self._load_inference_graph(inference_graph_path)
with self.test_session(graph=inference_graph):
inference_graph.get_tensor_by_name('image_tensor:0')
inference_graph.get_tensor_by_name('detection_boxes:0')
inference_graph.get_tensor_by_name('detection_scores:0')
inference_graph.get_tensor_by_name('detection_classes:0')
inference_graph.get_tensor_by_name('num_detections:0')
with self.assertRaises(KeyError):
inference_graph.get_tensor_by_name('detection_masks:0')
def test_export_and_run_inference_with_image_tensor(self):
checkpoint_path = os.path.join(self.get_temp_dir(), 'model-ckpt')
self._save_checkpoint_from_mock_model(checkpoint_path,
......@@ -161,7 +236,7 @@ class ExportInferenceGraphTest(tf.test.TestCase):
'exported_graph.pb')
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(num_classes=1)
mock_builder.return_value = FakeModel(add_detection_masks=True)
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
pipeline_config.eval_config.use_moving_averages = False
exporter.export_inference_graph(
......@@ -176,16 +251,72 @@ class ExportInferenceGraphTest(tf.test.TestCase):
boxes = inference_graph.get_tensor_by_name('detection_boxes:0')
scores = inference_graph.get_tensor_by_name('detection_scores:0')
classes = inference_graph.get_tensor_by_name('detection_classes:0')
masks = inference_graph.get_tensor_by_name('detection_masks:0')
num_detections = inference_graph.get_tensor_by_name('num_detections:0')
(boxes, scores, classes, num_detections) = sess.run(
[boxes, scores, classes, num_detections],
(boxes, scores, classes, masks, num_detections) = sess.run(
[boxes, scores, classes, masks, num_detections],
feed_dict={image_tensor: np.ones((1, 4, 4, 3)).astype(np.uint8)})
self.assertAllClose(boxes, [[0.0, 0.0, 0.5, 0.5],
[0.5, 0.5, 0.8, 0.8]])
self.assertAllClose(scores, [[0.7, 0.6]])
self.assertAllClose(classes, [[1, 2]])
self.assertAllClose(masks, np.arange(32).reshape([2, 4, 4]))
self.assertAllClose(num_detections, [2])
def _create_encoded_image_string(self, image_array_np, encoding_format):
od_graph = tf.Graph()
with od_graph.as_default():
if encoding_format == 'jpg':
encoded_string = tf.image.encode_jpeg(image_array_np)
elif encoding_format == 'png':
encoded_string = tf.image.encode_png(image_array_np)
else:
raise ValueError('Supports only the following formats: `jpg`, `png`')
with self.test_session(graph=od_graph):
return encoded_string.eval()
def test_export_and_run_inference_with_encoded_image_string_tensor(self):
checkpoint_path = os.path.join(self.get_temp_dir(), 'model-ckpt')
self._save_checkpoint_from_mock_model(checkpoint_path,
use_moving_averages=False)
inference_graph_path = os.path.join(self.get_temp_dir(),
'exported_graph.pb')
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(add_detection_masks=True)
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
pipeline_config.eval_config.use_moving_averages = False
exporter.export_inference_graph(
input_type='encoded_image_string_tensor',
pipeline_config=pipeline_config,
checkpoint_path=checkpoint_path,
inference_graph_path=inference_graph_path)
inference_graph = self._load_inference_graph(inference_graph_path)
jpg_image_str = self._create_encoded_image_string(
np.ones((4, 4, 3)).astype(np.uint8), 'jpg')
png_image_str = self._create_encoded_image_string(
np.ones((4, 4, 3)).astype(np.uint8), 'png')
with self.test_session(graph=inference_graph) as sess:
image_str_tensor = inference_graph.get_tensor_by_name(
'encoded_image_string_tensor:0')
boxes = inference_graph.get_tensor_by_name('detection_boxes:0')
scores = inference_graph.get_tensor_by_name('detection_scores:0')
classes = inference_graph.get_tensor_by_name('detection_classes:0')
masks = inference_graph.get_tensor_by_name('detection_masks:0')
num_detections = inference_graph.get_tensor_by_name('num_detections:0')
for image_str in [jpg_image_str, png_image_str]:
(boxes_np, scores_np, classes_np, masks_np,
num_detections_np) = sess.run(
[boxes, scores, classes, masks, num_detections],
feed_dict={image_str_tensor: image_str})
self.assertAllClose(boxes_np, [[0.0, 0.0, 0.5, 0.5],
[0.5, 0.5, 0.8, 0.8]])
self.assertAllClose(scores_np, [[0.7, 0.6]])
self.assertAllClose(classes_np, [[1, 2]])
self.assertAllClose(masks_np, np.arange(32).reshape([2, 4, 4]))
self.assertAllClose(num_detections_np, [2])
def test_export_and_run_inference_with_tf_example(self):
checkpoint_path = os.path.join(self.get_temp_dir(), 'model-ckpt')
self._save_checkpoint_from_mock_model(checkpoint_path,
......@@ -194,7 +325,7 @@ class ExportInferenceGraphTest(tf.test.TestCase):
'exported_graph.pb')
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(num_classes=1)
mock_builder.return_value = FakeModel(add_detection_masks=True)
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
pipeline_config.eval_config.use_moving_averages = False
exporter.export_inference_graph(
......@@ -209,17 +340,58 @@ class ExportInferenceGraphTest(tf.test.TestCase):
boxes = inference_graph.get_tensor_by_name('detection_boxes:0')
scores = inference_graph.get_tensor_by_name('detection_scores:0')
classes = inference_graph.get_tensor_by_name('detection_classes:0')
masks = inference_graph.get_tensor_by_name('detection_masks:0')
num_detections = inference_graph.get_tensor_by_name('num_detections:0')
(boxes, scores, classes, num_detections) = sess.run(
[boxes, scores, classes, num_detections],
(boxes, scores, classes, masks, num_detections) = sess.run(
[boxes, scores, classes, masks, num_detections],
feed_dict={tf_example: self._create_tf_example(
np.ones((4, 4, 3)).astype(np.uint8))})
self.assertAllClose(boxes, [[0.0, 0.0, 0.5, 0.5],
[0.5, 0.5, 0.8, 0.8]])
self.assertAllClose(scores, [[0.7, 0.6]])
self.assertAllClose(classes, [[1, 2]])
self.assertAllClose(masks, np.arange(32).reshape([2, 4, 4]))
self.assertAllClose(num_detections, [2])
def test_export_saved_model_and_run_inference(self):
checkpoint_path = os.path.join(self.get_temp_dir(), 'model-ckpt')
self._save_checkpoint_from_mock_model(checkpoint_path,
use_moving_averages=False)
inference_graph_path = os.path.join(self.get_temp_dir(),
'saved_model')
with mock.patch.object(
model_builder, 'build', autospec=True) as mock_builder:
mock_builder.return_value = FakeModel(add_detection_masks=True)
pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
pipeline_config.eval_config.use_moving_averages = False
exporter.export_inference_graph(
input_type='tf_example',
pipeline_config=pipeline_config,
checkpoint_path=checkpoint_path,
inference_graph_path=inference_graph_path,
export_as_saved_model=True)
with tf.Graph().as_default() as od_graph:
with self.test_session(graph=od_graph) as sess:
tf.saved_model.loader.load(
sess, [tf.saved_model.tag_constants.SERVING], inference_graph_path)
tf_example = od_graph.get_tensor_by_name('import/tf_example:0')
boxes = od_graph.get_tensor_by_name('import/detection_boxes:0')
scores = od_graph.get_tensor_by_name('import/detection_scores:0')
classes = od_graph.get_tensor_by_name('import/detection_classes:0')
masks = od_graph.get_tensor_by_name('import/detection_masks:0')
num_detections = od_graph.get_tensor_by_name('import/num_detections:0')
(boxes, scores, classes, masks, num_detections) = sess.run(
[boxes, scores, classes, masks, num_detections],
feed_dict={tf_example: self._create_tf_example(
np.ones((4, 4, 3)).astype(np.uint8))})
self.assertAllClose(boxes, [[0.0, 0.0, 0.5, 0.5],
[0.5, 0.5, 0.8, 0.8]])
self.assertAllClose(scores, [[0.7, 0.6]])
self.assertAllClose(classes, [[1, 2]])
self.assertAllClose(masks, np.arange(32).reshape([2, 4, 4]))
self.assertAllClose(num_detections, [2])
if __name__ == '__main__':
tf.test.main()
object_detection/g3doc/preparing_inputs.md
View file @ f906646c
......@@ -2,7 +2,7 @@
Tensorflow Object Detection API reads data using the TFRecord file format. Two
sample scripts (`create_pascal_tf_record.py` and `create_pet_tf_record.py`) are
provided to convert from the PASCAL VOC dataset and Oxford-IIT Pet dataset to
provided to convert from the PASCAL VOC dataset and Oxford-IIIT Pet dataset to
TFRecords.
## Generating the PASCAL VOC TFRecord files.
......@@ -11,35 +11,35 @@ The raw 2012 PASCAL VOC data set can be downloaded
[here](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar).
Extract the tar file and run the `create_pascal_tf_record` script:
```
```bash
# From tensorflow/models/object_detection
tar -xvf VOCtrainval_11-May-2012.tar
./create_pascal_tf_record --data_dir=VOCdevkit \
python create_pascal_tf_record.py --data_dir=VOCdevkit \
--year=VOC2012 --set=train --output_path=pascal_train.record
./create_pascal_tf_record --data_dir=/home/user/VOCdevkit \
python create_pascal_tf_record.py --data_dir=VOCdevkit \
--year=VOC2012 --set=val --output_path=pascal_val.record
```
You should end up with two TFRecord files named pascal_train.record and
pascal_val.record in the tensorflow/models/object_detection directory.
You should end up with two TFRecord files named `pascal_train.record` and
`pascal_val.record` in the `tensorflow/models/object_detection` directory.
The label map for the PASCAL VOC data set can be found at
data/pascal_label_map.pbtxt.
`data/pascal_label_map.pbtxt`.
## Generation the Oxford-IIT Pet TFRecord files.
## Generation the Oxford-IIIT Pet TFRecord files.
The Oxford-IIT Pet data set can be downloaded from
The Oxford-IIIT Pet data set can be downloaded from
[their website](http://www.robots.ox.ac.uk/~vgg/data/pets/). Extract the tar
file and run the `create_pet_tf_record` script to generate TFRecords.
```
```bash
# From tensorflow/models/object_detection
tar -xvf annotations.tar.gz
tar -xvf images.tar.gz
./create_pet_tf_record --data_dir=`pwd` --output_dir=`pwd`
python create_pet_tf_record.py --data_dir=`pwd` --output_dir=`pwd`
```
You should end up with two TFRecord files named pet_train.record and
pet_val.record in the tensorflow/models/object_detection directory.
You should end up with two TFRecord files named `pet_train.record` and
`pet_val.record` in the `tensorflow/models/object_detection` directory.
The label map for the Pet dataset can be found at data/pet_label_map.pbtxt.
The label map for the Pet dataset can be found at `data/pet_label_map.pbtxt`.
object_detection/g3doc/running_locally.md
View file @ f906646c
......@@ -10,7 +10,7 @@ dependencies, compiling the configuration protobufs and setting up the Python
environment.
2. A valid data set has been created. See [this page](preparing_inputs.md) for
instructions on how to generate a dataset for the PASCAL VOC challenge or the
Oxford-IIT Pet dataset.
Oxford-IIIT Pet dataset.
3. A Object Detection pipeline configuration has been written. See
[this page](configuring_jobs.md) for details on how to write a pipeline configuration.
......
object_detection/g3doc/running_on_cloud.md
View file @ f906646c
......@@ -11,7 +11,7 @@ See [the Cloud ML quick start guide](https://cloud.google.com/ml-engine/docs/qui
in the [installation instructions](installation.md).
3. The reader has a valid data set and stored it in a Google Cloud Storage
bucket. See [this page](preparing_inputs.md) for instructions on how to generate
a dataset for the PASCAL VOC challenge or the Oxford-IIT Pet dataset.
a dataset for the PASCAL VOC challenge or the Oxford-IIIT Pet dataset.
4. The reader has configured a valid Object Detection pipeline, and stored it
in a Google Cloud Storage bucket. See [this page](configuring_jobs.md) for
details on how to write a pipeline configuration.
......@@ -88,7 +88,7 @@ training checkpoints and events will be written to and
Google Cloud Storage.
Users can monitor the progress of their training job on the [ML Engine
Dasboard](https://pantheon.corp.google.com/mlengine/jobs).
Dashboard](https://pantheon.corp.google.com/mlengine/jobs).
## Running an Evaluation Job on Cloud
......
object_detection/g3doc/running_pets.md
View file @ f906646c
# Quick Start: Distributed Training on the Oxford-IIT Pets Dataset on Google Cloud
# Quick Start: Distributed Training on the Oxford-IIIT Pets Dataset on Google Cloud
This page is a walkthrough for training an object detector using the Tensorflow
Object Detection API. In this tutorial, we'll be training on the Oxford-IIT Pets
Object Detection API. In this tutorial, we'll be training on the Oxford-IIIT Pets
dataset to build a system to detect various breeds of cats and dogs. The output
of the detector will look like the following:
......@@ -43,16 +43,16 @@ Please run through the [installation instructions](installation.md) to install
Tensorflow and all it dependencies. Ensure the Protobuf libraries are
compiled and the library directories are added to `PYTHONPATH`.
## Getting the Oxford-IIT Pets Dataset and Uploading it to Google Cloud Storage
## Getting the Oxford-IIIT Pets Dataset and Uploading it to Google Cloud Storage
In order to train a detector, we require a dataset of images, bounding boxes and
classifications. For this demo, we'll use the Oxford-IIT Pets dataset. The raw
dataset for Oxford-IIT Pets lives
classifications. For this demo, we'll use the Oxford-IIIT Pets dataset. The raw
dataset for Oxford-IIIT Pets lives
[here](http://www.robots.ox.ac.uk/~vgg/data/pets/). You will need to download
both the image dataset [`images.tar.gz`](http://www.robots.ox.ac.uk/~vgg/data/pets/data/images.tar.gz)
and the groundtruth data [`annotations.tar.gz`](http://www.robots.ox.ac.uk/~vgg/data/pets/data/annotations.tar.gz)
to the tensorflow/models directory. This may take some time. After downloading
the tarballs, your object_detection directory should appear as follows:
to the `tensorflow/models` directory. This may take some time. After downloading
the tarballs, your `object_detection` directory should appear as follows:
```lang-none
+ object_detection/
......@@ -64,9 +64,9 @@ the tarballs, your object_detection directory should appear as follows:
```
The Tensorflow Object Detection API expects data to be in the TFRecord format,
so we'll now run the _create_pet_tf_record_ script to convert from the raw
Oxford-IIT Pet dataset into TFRecords. Run the following commands from the
object_detection directory:
so we'll now run the `create_pet_tf_record` script to convert from the raw
Oxford-IIIT Pet dataset into TFRecords. Run the following commands from the
`object_detection` directory:
``` bash
# From tensorflow/models/
......@@ -83,12 +83,12 @@ python object_detection/create_pet_tf_record.py \
Note: It is normal to see some warnings when running this script. You may ignore
them.
Two TFRecord files named pet_train.record and pet_val.record should be generated
in the object_detection/ directory.
Two TFRecord files named `pet_train.record` and `pet_val.record` should be generated
in the `object_detection` directory.
Now that the data has been generated, we'll need to upload it to Google Cloud
Storage so the data can be accessed by ML Engine. Run the following command to
copy the files into your GCS bucket (substituting ${YOUR_GCS_BUCKET}):
copy the files into your GCS bucket (substituting `${YOUR_GCS_BUCKET}`):
``` bash
# From tensorflow/models/
......@@ -109,7 +109,7 @@ parameters to initialize our new model.
Download our [COCO-pretrained Faster R-CNN with Resnet-101
model](http://storage.googleapis.com/download.tensorflow.org/models/object_detection/faster_rcnn_resnet101_coco_11_06_2017.tar.gz).
Unzip the contents of the folder and copy the model.ckpt* files into your GCS
Unzip the contents of the folder and copy the `model.ckpt*` files into your GCS
Bucket.
``` bash
......@@ -127,14 +127,14 @@ In the Tensorflow Object Detection API, the model parameters, training
parameters and eval parameters are all defined by a config file. More details
can be found [here](configuring_jobs.md). For this tutorial, we will use some
predefined templates provided with the source code. In the
object_detection/samples/configs folder, there are skeleton object_detection
`object_detection/samples/configs` folder, there are skeleton object_detection
configuration files. We will use `faster_rcnn_resnet101_pets.config` as a
starting point for configuring the pipeline. Open the file with your favourite
text editor.
We'll need to configure some paths in order for the template to work. Search the
file for instances of `PATH_TO_BE_CONFIGURED` and replace them with the
appropriate value (typically "gs://${YOUR_GCS_BUCKET}/data/"). Afterwards
appropriate value (typically `gs://${YOUR_GCS_BUCKET}/data/`). Afterwards
upload your edited file onto GCS, making note of the path it was uploaded to
(we'll need it when starting the training/eval jobs).
......@@ -146,7 +146,7 @@ upload your edited file onto GCS, making note of the path it was uploaded to
sed -i "s|PATH_TO_BE_CONFIGURED|"gs://${YOUR_GCS_BUCKET}"/data|g" \
object_detection/samples/configs/faster_rcnn_resnet101_pets.config
# Copy editted template to cloud.
# Copy edited template to cloud.
gsutil cp object_detection/samples/configs/faster_rcnn_resnet101_pets.config \
gs://${YOUR_GCS_BUCKET}/data/faster_rcnn_resnet101_pets.config
```
......@@ -171,7 +171,7 @@ the following:
```
You can inspect your bucket using the [Google Cloud Storage
browser](pantheon.corp.google.com/storage).
browser](https://console.cloud.google.com/storage/browser).
## Starting Training and Evaluation Jobs on Google Cloud ML Engine
......@@ -181,7 +181,7 @@ Before we can start a job on Google Cloud ML Engine, we must:
2. Write a cluster configuration for our Google Cloud ML job.
To package the Tensorflow Object Detection code, run the following commands from
the tensorflow/models/ directory:
the `tensorflow/models/` directory:
``` bash
# From tensorflow/models/
......@@ -194,9 +194,9 @@ and `slim/dist/slim-0.1.tar.gz`.
For running the training Cloud ML job, we'll configure the cluster to use 10
training jobs (1 master + 9 workers) and three parameters servers. The
configuration file can be found at object_detection/samples/cloud/cloud.yml.
configuration file can be found at `object_detection/samples/cloud/cloud.yml`.
To start training, execute the following command from the tensorflow/models/
To start training, execute the following command from the `tensorflow/models/`
directory:
``` bash
......@@ -233,7 +233,7 @@ submit training` command is correct. ML Engine does not distinguish between
training and evaluation jobs.
Users can monitor and stop training and evaluation jobs on the [ML Engine
Dasboard](https://pantheon.corp.google.com/mlengine/jobs).
Dashboard](https://console.cloud.google.com/mlengine/jobs).
## Monitoring Progress with Tensorboard
......@@ -263,35 +263,35 @@ Note: It takes roughly 10 minutes for a job to get started on ML Engine, and
roughly an hour for the system to evaluate the validation dataset. It may take
some time to populate the dashboards. If you do not see any entries after half
an hour, check the logs from the [ML Engine
Dasboard](https://pantheon.corp.google.com/mlengine/jobs).
Dashboard](https://console.cloud.google.com/mlengine/jobs).
## Exporting the Tensorflow Graph
After your model has been trained, you should export it to a Tensorflow
graph proto. First, you need to identify a candidate checkpoint to export. You
can search your bucket using the [Google Cloud Storage
Browser](https://pantheon.corp.google.com/storage/browser). The file should be
stored under ${YOUR_GCS_BUCKET}/train. The checkpoint will typically consist of
Browser](https://console.cloud.google.com/storage/browser). The file should be
stored under `${YOUR_GCS_BUCKET}/train`. The checkpoint will typically consist of
three files:
* model.ckpt-${CHECKPOINT_NUMBER}.data-00000-of-00001,
* model.ckpt-${CHECKPOINT_NUMBER}.index
* model.ckpt-${CHECKPOINT_NUMBER}.meta
* `model.ckpt-${CHECKPOINT_NUMBER}.data-00000-of-00001`
* `model.ckpt-${CHECKPOINT_NUMBER}.index`
* `model.ckpt-${CHECKPOINT_NUMBER}.meta`
After you've identified a candidate checkpoint to export, run the following
command from tensorflow/models/object_detection:
command from `tensorflow/models/object_detection`:
``` bash
# From tensorflow/models
gsutil cp gs://${YOUR_GCS_BUCKET}/train/model.ckpt-${CHECKPOINT_NUMBER}.* .
python object_detection/export_inference_graph \
python object_detection/export_inference_graph.py \
--input_type image_tensor \
--pipeline_config_path object_detection/samples/configs/faster_rcnn_resnet101_pets.config \
--checkpoint_path model.ckpt-${CHECKPOINT_NUMBER} \
--inference_graph_path output_inference_graph.pb
```
Afterwards, you should see a graph named output_inference_graph.pb.
Afterwards, you should see a graph named `output_inference_graph.pb`.
## What's Next
......
object_detection/samples/configs/faster_rcnn_inception_resnet_v2_atrous_pets.config
View file @ f906646c
# Faster R-CNN with Inception Resnet v2, Atrous version;
# Configured for Oxford-IIT Pets Dataset.
# Configured for Oxford-IIIT Pets Dataset.
# Users should configure the fine_tune_checkpoint field in the train config as
# well as the label_map_path and input_path fields in the train_input_reader and
# eval_input_reader. Search for "PATH_TO_BE_CONFIGURED" to find the fields that
......
object_detection/samples/configs/faster_rcnn_resnet101_pets.config
View file @ f906646c
# Faster R-CNN with Resnet-101 (v1) configured for the Oxford-IIT Pet Dataset.
# Faster R-CNN with Resnet-101 (v1) configured for the Oxford-IIIT Pet Dataset.
# Users should configure the fine_tune_checkpoint field in the train config as
# well as the label_map_path and input_path fields in the train_input_reader and
# eval_input_reader. Search for "PATH_TO_BE_CONFIGURED" to find the fields that
......
object_detection/samples/configs/faster_rcnn_resnet101_voc07.config
View file @ f906646c
......@@ -118,9 +118,9 @@ train_config: {
train_input_reader: {
tf_record_input_reader {
input_path: "PATH_TO_BE_CONFIGURED/pascal_voc_train.record"
input_path: "PATH_TO_BE_CONFIGURED/pascal_train.record"
}
label_map_path: "PATH_TO_BE_CONFIGURED/pascal_voc_label_map.pbtxt"
label_map_path: "PATH_TO_BE_CONFIGURED/pascal_label_map.pbtxt"
}
eval_config: {
......@@ -129,7 +129,7 @@ eval_config: {
eval_input_reader: {
tf_record_input_reader {
input_path: "PATH_TO_BE_CONFIGURED/pascal_voc_val.record"
input_path: "PATH_TO_BE_CONFIGURED/pascal_val.record"
}
label_map_path: "PATH_TO_BE_CONFIGURED/pascal_voc_label_map.pbtxt"
label_map_path: "PATH_TO_BE_CONFIGURED/pascal_label_map.pbtxt"
}
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment