Skip to content

GitLab

"docs/vscode:/vscode.git/clone" did not exist on "d17986f8c6b50e8acc97ad8fc0a965d8ef70a30e"
Commit 78d5f8f8 authored by Zhichao Lu's avatar Zhichao Lu Committed by lzc5123016
Browse files

Merged commit includes the following changes: 

187187978  by Zhichao Lu:

    Only updating hyperparameters if they have non-null values.

--
187097690  by Zhichao Lu:

    Rewrite some conditions a bit more clearly.

--
187085190  by Zhichao Lu:

    More informative error message.

--
186935376  by Zhichao Lu:

    Added option to evaluator.evaluate to use custom evaluator objects.

--
186808249  by Zhichao Lu:

    Fix documentation re: number of stages.

--
186775014  by Zhichao Lu:

    Change anchor generator interface to return a list of BoxLists containing anchors for different feature map layers.

--
186729028  by Zhichao Lu:

    Minor fixes to object detection.

--
186723716  by Zhichao Lu:

    Fix tf_example_decoder.py initailization issue.

--
186668505  by Zhichao Lu:

    Remove unused import.

--
186475361  by Zhichao Lu:

    Update the box predictor interface to return list of predictions - one from each feature map - instead of stacking them into one large tensor.

--
186410844  by Zhich...
parent 629adffa
Hide whitespace changes
Inline Side-by-side
Showing with 433 additions and 719 deletions
research/object_detection/core/BUILD deleted 100644 → 0
View file @ 629adffa
# Tensorflow Object Detection API: Core.
package(
default_visibility = ["//visibility:public"],
)
licenses(["notice"])
# Apache 2.0
py_library(
name = "batcher",
srcs = ["batcher.py"],
deps = [
":prefetcher",
":preprocessor",
":standard_fields",
"//tensorflow",
],
)
py_test(
name = "batcher_test",
srcs = ["batcher_test.py"],
deps = [
":batcher",
"//tensorflow",
],
)
py_library(
name = "box_list",
srcs = [
"box_list.py",
],
deps = [
"//tensorflow",
],
)
py_test(
name = "box_list_test",
srcs = ["box_list_test.py"],
deps = [
":box_list",
],
)
py_library(
name = "box_list_ops",
srcs = [
"box_list_ops.py",
],
deps = [
":box_list",
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:shape_utils",
],
)
py_test(
name = "box_list_ops_test",
srcs = ["box_list_ops_test.py"],
deps = [
":box_list",
":box_list_ops",
],
)
py_library(
name = "box_coder",
srcs = [
"box_coder.py",
],
deps = [
"//tensorflow",
],
)
py_test(
name = "box_coder_test",
srcs = [
"box_coder_test.py",
],
deps = [
":box_coder",
":box_list",
"//tensorflow",
],
)
py_library(
name = "keypoint_ops",
srcs = [
"keypoint_ops.py",
],
deps = [
"//tensorflow",
],
)
py_test(
name = "keypoint_ops_test",
srcs = ["keypoint_ops_test.py"],
deps = [
":keypoint_ops",
],
)
py_library(
name = "losses",
srcs = ["losses.py"],
deps = [
":box_list",
":box_list_ops",
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:ops",
],
)
py_library(
name = "matcher",
srcs = [
"matcher.py",
],
deps = [
"//tensorflow/models/research/object_detection/utils:ops",
],
)
py_library(
name = "model",
srcs = ["model.py"],
deps = [
":standard_fields",
],
)
py_test(
name = "matcher_test",
srcs = [
"matcher_test.py",
],
deps = [
":matcher",
"//tensorflow",
],
)
py_library(
name = "prefetcher",
srcs = ["prefetcher.py"],
deps = ["//tensorflow"],
)
py_library(
name = "preprocessor",
srcs = [
"preprocessor.py",
],
deps = [
":box_list",
":box_list_ops",
":keypoint_ops",
":preprocessor_cache",
":standard_fields",
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:shape_utils",
],
)
py_library(
name = "preprocessor_cache",
srcs = [
"preprocessor_cache.py",
],
)
py_test(
name = "preprocessor_test",
srcs = [
"preprocessor_test.py",
],
deps = [
":preprocessor",
":preprocessor_cache",
"//tensorflow",
],
)
py_test(
name = "losses_test",
srcs = ["losses_test.py"],
deps = [
":box_list",
":losses",
":matcher",
"//tensorflow",
],
)
py_test(
name = "prefetcher_test",
srcs = ["prefetcher_test.py"],
deps = [
":prefetcher",
"//tensorflow",
],
)
py_library(
name = "standard_fields",
srcs = [
"standard_fields.py",
],
)
py_library(
name = "post_processing",
srcs = ["post_processing.py"],
deps = [
":box_list",
":box_list_ops",
":standard_fields",
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:shape_utils",
],
)
py_test(
name = "post_processing_test",
srcs = ["post_processing_test.py"],
deps = [
":box_list",
":box_list_ops",
":post_processing",
"//tensorflow",
],
)
py_library(
name = "target_assigner",
srcs = [
"target_assigner.py",
],
deps = [
":box_list",
":matcher",
":region_similarity_calculator",
":standard_fields",
"//tensorflow",
"//tensorflow/models/research/object_detection/box_coders:faster_rcnn_box_coder",
"//tensorflow/models/research/object_detection/box_coders:mean_stddev_box_coder",
"//tensorflow/models/research/object_detection/core:box_coder",
"//tensorflow/models/research/object_detection/matchers:argmax_matcher",
"//tensorflow/models/research/object_detection/matchers:bipartite_matcher",
"//tensorflow/models/research/object_detection/utils:shape_utils",
],
)
py_test(
name = "target_assigner_test",
size = "large",
timeout = "long",
srcs = ["target_assigner_test.py"],
deps = [
":box_list",
":region_similarity_calculator",
":target_assigner",
"//tensorflow",
"//tensorflow/models/research/object_detection/box_coders:keypoint_box_coder",
"//tensorflow/models/research/object_detection/box_coders:mean_stddev_box_coder",
"//tensorflow/models/research/object_detection/matchers:bipartite_matcher",
"//tensorflow/models/research/object_detection/utils:test_case",
],
)
py_library(
name = "data_decoder",
srcs = ["data_decoder.py"],
)
py_library(
name = "data_parser",
srcs = ["data_parser.py"],
)
py_library(
name = "box_predictor",
srcs = ["box_predictor.py"],
deps = [
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:ops",
"//tensorflow/models/research/object_detection/utils:shape_utils",
"//tensorflow/models/research/object_detection/utils:static_shape",
],
)
py_test(
name = "box_predictor_test",
srcs = ["box_predictor_test.py"],
deps = [
":box_predictor",
"//tensorflow",
"//tensorflow/models/research/object_detection/builders:hyperparams_builder",
"//tensorflow/models/research/object_detection/protos:hyperparams_py_pb2",
"//tensorflow/models/research/object_detection/utils:test_case",
],
)
py_library(
name = "region_similarity_calculator",
srcs = [
"region_similarity_calculator.py",
],
deps = [
"//tensorflow",
"//tensorflow/models/research/object_detection/core:box_list_ops",
],
)
py_test(
name = "region_similarity_calculator_test",
srcs = [
"region_similarity_calculator_test.py",
],
deps = [
":region_similarity_calculator",
"//tensorflow/models/research/object_detection/core:box_list",
],
)
py_library(
name = "anchor_generator",
srcs = [
"anchor_generator.py",
],
deps = [
"//tensorflow",
],
)
py_library(
name = "minibatch_sampler",
srcs = [
"minibatch_sampler.py",
],
deps = [
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:ops",
],
)
py_test(
name = "minibatch_sampler_test",
srcs = [
"minibatch_sampler_test.py",
],
deps = [
":minibatch_sampler",
"//tensorflow",
],
)
py_library(
name = "balanced_positive_negative_sampler",
srcs = [
"balanced_positive_negative_sampler.py",
],
deps = [
":minibatch_sampler",
"//tensorflow",
],
)
py_test(
name = "balanced_positive_negative_sampler_test",
srcs = [
"balanced_positive_negative_sampler_test.py",
],
deps = [
":balanced_positive_negative_sampler",
"//tensorflow",
],
)
research/object_detection/core/anchor_generator.py
View file @ 78d5f8f8
......@@ -77,7 +77,7 @@ class AnchorGenerator(object):
def generate(self, feature_map_shape_list, **params):
"""Generates a collection of bounding boxes to be used as anchors.
TODO: remove **params from argument list and make stride and
TODO(rathodv): remove **params from argument list and make stride and
offsets (for multiple_grid_anchor_generator) constructor arguments.
Args:
......@@ -88,7 +88,9 @@ class AnchorGenerator(object):
**params: parameters for anchor generation op
Returns:
boxes: a BoxList holding a collection of N anchor boxes
boxes_list: a list of BoxLists each holding anchor boxes corresponding to
the input feature map shapes.
Raises:
ValueError: if the number of feature map shapes does not match the length
of NumAnchorsPerLocation.
......@@ -98,13 +100,14 @@ class AnchorGenerator(object):
raise ValueError('Number of feature maps is expected to equal the length '
'of `num_anchors_per_location`.')
with tf.name_scope(self.name_scope()):
anchors = self._generate(feature_map_shape_list, **params)
anchors_list = self._generate(feature_map_shape_list, **params)
if self.check_num_anchors:
with tf.control_dependencies([
self._assert_correct_number_of_anchors(
anchors, feature_map_shape_list)]):
anchors.set(tf.identity(anchors.get()))
return anchors
anchors_list, feature_map_shape_list)]):
for item in anchors_list:
item.set(tf.identity(item.get()))
return anchors_list
@abstractmethod
def _generate(self, feature_map_shape_list, **params):
......@@ -117,15 +120,17 @@ class AnchorGenerator(object):
**params: parameters for anchor generation op
Returns:
boxes: a BoxList holding a collection of N anchor boxes
boxes_list: a list of BoxList, each holding a collection of N anchor
boxes.
"""
pass
def _assert_correct_number_of_anchors(self, anchors, feature_map_shape_list):
def _assert_correct_number_of_anchors(self, anchors_list,
feature_map_shape_list):
"""Assert that correct number of anchors was generated.
Args:
anchors: box_list.BoxList object holding anchors generated
anchors_list: A list of box_list.BoxList object holding anchors generated.
feature_map_shape_list: list of (height, width) pairs in the format
[(height_0, width_0), (height_1, width_1), ...] that the generated
anchors must align with.
......@@ -134,10 +139,12 @@ class AnchorGenerator(object):
match the number of expected anchors.
"""
expected_num_anchors = 0
for num_anchors_per_location, feature_map_shape in zip(
self.num_anchors_per_location(), feature_map_shape_list):
actual_num_anchors = 0
for num_anchors_per_location, feature_map_shape, anchors in zip(
self.num_anchors_per_location(), feature_map_shape_list, anchors_list):
expected_num_anchors += (num_anchors_per_location
* feature_map_shape[0]
* feature_map_shape[1])
return tf.assert_equal(expected_num_anchors, anchors.num_boxes())
actual_num_anchors += anchors.num_boxes()
return tf.assert_equal(expected_num_anchors, actual_num_anchors)
research/object_detection/core/box_list_ops.py
View file @ 78d5f8f8
......@@ -585,7 +585,7 @@ def sort_by_field(boxlist, field, order=SortOrder.descend, scope=None):
['Incorrect field size: actual vs expected.', num_entries, num_boxes])
with tf.control_dependencies([length_assert]):
# TODO: Remove with tf.device when top_k operation runs
# TODO(derekjchow): Remove with tf.device when top_k operation runs
# correctly on GPU.
with tf.device('/cpu:0'):
_, sorted_indices = tf.nn.top_k(field_to_sort, num_boxes, sorted=True)
......@@ -657,7 +657,7 @@ def filter_greater_than(boxlist, thresh, scope=None):
This op keeps the collection of boxes whose corresponding scores are
greater than the input threshold.
TODO: Change function name to filter_scores_greater_than
TODO(jonathanhuang): Change function name to filter_scores_greater_than
Args:
boxlist: BoxList holding N boxes. Must contain a 'scores' field
......@@ -937,7 +937,7 @@ def box_voting(selected_boxes, pool_boxes, iou_thresh=0.5):
iou_ = iou(selected_boxes, pool_boxes)
match_indicator = tf.to_float(tf.greater(iou_, iou_thresh))
num_matches = tf.reduce_sum(match_indicator, 1)
# TODO: Handle the case where some boxes in selected_boxes do not
# TODO(kbanoop): Handle the case where some boxes in selected_boxes do not
# match to any boxes in pool_boxes. For such boxes without any matches, we
# should return the original boxes without voting.
match_assert = tf.Assert(
......
research/object_detection/core/box_predictor.py
View file @ 78d5f8f8
......@@ -64,11 +64,9 @@ class BoxPredictor(object):
scope=None, **params):
"""Computes encoded object locations and corresponding confidences.
Takes a high level image feature map as input and produce two predictions,
(1) a tensor encoding box locations, and
(2) a tensor encoding class scores for each corresponding box.
In this interface, we only assume that two tensors are returned as output
and do not assume anything about their shapes.
Takes a list of high level image feature maps as input and produces a list
of box encodings and a list of class scores where each element in the output
lists correspond to the feature maps in the input list.
Args:
image_features: A list of float tensors of shape [batch_size, height_i,
......@@ -81,12 +79,14 @@ class BoxPredictor(object):
Returns:
A dictionary containing at least the following tensors.
box_encodings: A float tensor of shape
[batch_size, num_anchors, q, code_size] representing the location of
the objects, where q is 1 or the number of classes.
class_predictions_with_background: A float tensor of shape
[batch_size, num_anchors, num_classes + 1] representing the class
predictions for the proposals.
box_encodings: A list of float tensors of shape
[batch_size, num_anchors_i, q, code_size] representing the location of
the objects, where q is 1 or the number of classes. Each entry in the
list corresponds to a feature map in the input `image_features` list.
class_predictions_with_background: A list of float tensors of shape
[batch_size, num_anchors_i, num_classes + 1] representing the class
predictions for the proposals. Each entry in the list corresponds to a
feature map in the input `image_features` list.
Raises:
ValueError: If length of `image_features` is not equal to length of
......@@ -104,7 +104,7 @@ class BoxPredictor(object):
return self._predict(image_features, num_predictions_per_location,
**params)
# TODO: num_predictions_per_location could be moved to constructor.
# TODO(rathodv): num_predictions_per_location could be moved to constructor.
# This is currently only used by ConvolutionalBoxPredictor.
@abstractmethod
def _predict(self, image_features, num_predictions_per_location, **params):
......@@ -120,12 +120,14 @@ class BoxPredictor(object):
Returns:
A dictionary containing at least the following tensors.
box_encodings: A float tensor of shape
[batch_size, num_anchors, q, code_size] representing the location of
the objects, where q is 1 or the number of classes.
class_predictions_with_background: A float tensor of shape
[batch_size, num_anchors, num_classes + 1] representing the class
predictions for the proposals.
box_encodings: A list of float tensors of shape
[batch_size, num_anchors_i, q, code_size] representing the location of
the objects, where q is 1 or the number of classes. Each entry in the
list corresponds to a feature map in the input `image_features` list.
class_predictions_with_background: A list of float tensors of shape
[batch_size, num_anchors_i, num_classes + 1] representing the class
predictions for the proposals. Each entry in the list corresponds to a
feature map in the input `image_features` list.
"""
pass
......@@ -133,7 +135,7 @@ class BoxPredictor(object):
class RfcnBoxPredictor(BoxPredictor):
"""RFCN Box Predictor.
Applies a position sensitve ROI pooling on position sensitive feature maps to
Applies a position sensitive ROI pooling on position sensitive feature maps to
predict classes and refined locations. See https://arxiv.org/abs/1605.06409
for details.
......@@ -191,12 +193,14 @@ class RfcnBoxPredictor(BoxPredictor):
box_code_size].
Returns:
box_encodings: A float tensor of shape
[batch_size, num_anchors, num_classes, code_size] representing the
location of the objects.
class_predictions_with_background: A float tensor of shape
[batch_size, num_anchors, num_classes + 1] representing the class
predictions for the proposals.
box_encodings: A list of float tensors of shape
[batch_size, num_anchors_i, q, code_size] representing the location of
the objects, where q is 1 or the number of classes. Each entry in the
list corresponds to a feature map in the input `image_features` list.
class_predictions_with_background: A list of float tensors of shape
[batch_size, num_anchors_i, num_classes + 1] representing the class
predictions for the proposals. Each entry in the list corresponds to a
feature map in the input `image_features` list.
Raises:
ValueError: if num_predictions_per_location is not 1 or if
......@@ -266,11 +270,12 @@ class RfcnBoxPredictor(BoxPredictor):
class_predictions_with_background,
[batch_size * num_boxes, 1, total_classes])
return {BOX_ENCODINGS: box_encodings,
return {BOX_ENCODINGS: [box_encodings],
CLASS_PREDICTIONS_WITH_BACKGROUND:
class_predictions_with_background}
[class_predictions_with_background]}
# TODO(rathodv): Change the implementation to return lists of predictions.
class MaskRCNNBoxPredictor(BoxPredictor):
"""Mask R-CNN Box Predictor.
......@@ -644,18 +649,18 @@ class ConvolutionalBoxPredictor(BoxPredictor):
feature map.
Returns:
A dictionary containing the following tensors.
box_encodings: A float tensor of shape [batch_size, num_anchors, 1,
code_size] representing the location of the objects, where
num_anchors = feat_height * feat_width * num_predictions_per_location
class_predictions_with_background: A float tensor of shape
[batch_size, num_anchors, num_classes + 1] representing the class
predictions for the proposals.
box_encodings: A list of float tensors of shape
[batch_size, num_anchors_i, q, code_size] representing the location of
the objects, where q is 1 or the number of classes. Each entry in the
list corresponds to a feature map in the input `image_features` list.
class_predictions_with_background: A list of float tensors of shape
[batch_size, num_anchors_i, num_classes + 1] representing the class
predictions for the proposals. Each entry in the list corresponds to a
feature map in the input `image_features` list.
"""
box_encodings_list = []
class_predictions_list = []
# TODO: Come up with a better way to generate scope names
# TODO(rathodv): Come up with a better way to generate scope names
# in box predictor once we have time to retrain all models in the zoo.
# The following lines create scope names to be backwards compatible with the
# existing checkpoints.
......@@ -741,12 +746,13 @@ class ConvolutionalBoxPredictor(BoxPredictor):
num_predictions_per_location,
num_class_slots]))
class_predictions_list.append(class_predictions_with_background)
return {BOX_ENCODINGS: tf.concat(box_encodings_list, axis=1),
CLASS_PREDICTIONS_WITH_BACKGROUND:
tf.concat(class_predictions_list, axis=1)}
return {
BOX_ENCODINGS: box_encodings_list,
CLASS_PREDICTIONS_WITH_BACKGROUND: class_predictions_list
}
# TODO: Merge the implementation with ConvolutionalBoxPredictor above
# TODO(rathodv): Merge the implementation with ConvolutionalBoxPredictor above
# since they are very similar.
class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
"""Convolutional Box Predictor with weight sharing.
......@@ -806,13 +812,14 @@ class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
shared.
Returns:
A dictionary containing the following tensors.
box_encodings: A float tensor of shape [batch_size, num_anchors, 1,
code_size] representing the location of the objects, where
num_anchors = feat_height * feat_width * num_predictions_per_location
class_predictions_with_background: A float tensor of shape
[batch_size, num_anchors, num_classes + 1] representing the class
predictions for the proposals.
box_encodings: A list of float tensors of shape
[batch_size, num_anchors_i, q, code_size] representing the location of
the objects, where q is 1 or the number of classes. Each entry in the
list corresponds to a feature map in the input `image_features` list.
class_predictions_with_background: A list of float tensors of shape
[batch_size, num_anchors_i, num_classes + 1] representing the class
predictions for the proposals. Each entry in the list corresponds to a
feature map in the input `image_features` list.
Raises:
ValueError: If the image feature maps do not have the same number of
......@@ -890,6 +897,7 @@ class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
num_predictions_per_location,
num_class_slots]))
class_predictions_list.append(class_predictions_with_background)
return {BOX_ENCODINGS: tf.concat(box_encodings_list, axis=1),
CLASS_PREDICTIONS_WITH_BACKGROUND:
tf.concat(class_predictions_list, axis=1)}
return {
BOX_ENCODINGS: box_encodings_list,
CLASS_PREDICTIONS_WITH_BACKGROUND: class_predictions_list
}
research/object_detection/core/box_predictor_test.py
View file @ 78d5f8f8
......@@ -165,9 +165,11 @@ class RfcnBoxPredictorTest(tf.test.TestCase):
[image_features], num_predictions_per_location=[1],
scope='BoxPredictor',
proposal_boxes=proposal_boxes)
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
class_predictions_with_background = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
class_predictions_with_background = tf.concat(
box_predictions[box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND],
axis=1)
init_op = tf.global_variables_initializer()
with self.test_session() as sess:
......@@ -215,9 +217,11 @@ class ConvolutionalBoxPredictorTest(test_case.TestCase):
box_predictions = conv_box_predictor.predict(
[image_features], num_predictions_per_location=[5],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
objectness_predictions = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
objectness_predictions = tf.concat(
box_predictions[box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND],
axis=1)
return (box_encodings, objectness_predictions)
image_features = np.random.rand(4, 8, 8, 64).astype(np.float32)
(box_encodings, objectness_predictions) = self.execute(graph_fn,
......@@ -242,9 +246,10 @@ class ConvolutionalBoxPredictorTest(test_case.TestCase):
box_predictions = conv_box_predictor.predict(
[image_features], num_predictions_per_location=[1],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
objectness_predictions = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
objectness_predictions = tf.concat(box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND], axis=1)
return (box_encodings, objectness_predictions)
image_features = np.random.rand(4, 8, 8, 64).astype(np.float32)
(box_encodings, objectness_predictions) = self.execute(graph_fn,
......@@ -273,9 +278,11 @@ class ConvolutionalBoxPredictorTest(test_case.TestCase):
[image_features],
num_predictions_per_location=[5],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
class_predictions_with_background = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
class_predictions_with_background = tf.concat(
box_predictions[box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND],
axis=1)
return (box_encodings, class_predictions_with_background)
(box_encodings,
class_predictions_with_background) = self.execute(graph_fn,
......@@ -302,9 +309,11 @@ class ConvolutionalBoxPredictorTest(test_case.TestCase):
box_predictions = conv_box_predictor.predict(
[image_features], num_predictions_per_location=[5],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
objectness_predictions = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
objectness_predictions = tf.concat(
box_predictions[box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND],
axis=1)
init_op = tf.global_variables_initializer()
resolution = 32
......@@ -348,9 +357,11 @@ class ConvolutionalBoxPredictorTest(test_case.TestCase):
box_predictions = conv_box_predictor.predict(
[image_features], num_predictions_per_location=[5],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
objectness_predictions = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
objectness_predictions = tf.concat(
box_predictions[box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND],
axis=1)
init_op = tf.global_variables_initializer()
resolution = 32
......@@ -412,9 +423,10 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
box_predictions = conv_box_predictor.predict(
[image_features], num_predictions_per_location=[5],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
objectness_predictions = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
objectness_predictions = tf.concat(box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND], axis=1)
return (box_encodings, objectness_predictions)
image_features = np.random.rand(4, 8, 8, 64).astype(np.float32)
(box_encodings, objectness_predictions) = self.execute(
......@@ -438,9 +450,10 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
[image_features],
num_predictions_per_location=[5],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
class_predictions_with_background = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
class_predictions_with_background = tf.concat(box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND], axis=1)
return (box_encodings, class_predictions_with_background)
image_features = np.random.rand(4, 8, 8, 64).astype(np.float32)
......@@ -466,9 +479,11 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
[image_features1, image_features2],
num_predictions_per_location=[5, 5],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
class_predictions_with_background = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
class_predictions_with_background = tf.concat(
box_predictions[box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND],
axis=1)
return (box_encodings, class_predictions_with_background)
image_features1 = np.random.rand(4, 8, 8, 64).astype(np.float32)
......@@ -493,9 +508,11 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
[image_features1, image_features2],
num_predictions_per_location=[5, 5],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
class_predictions_with_background = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
class_predictions_with_background = tf.concat(
box_predictions[box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND],
axis=1)
return (box_encodings, class_predictions_with_background)
with self.test_session(graph=tf.Graph()):
......@@ -543,9 +560,10 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
box_predictions = conv_box_predictor.predict(
[image_features], num_predictions_per_location=[5],
scope='BoxPredictor')
box_encodings = box_predictions[box_predictor.BOX_ENCODINGS]
objectness_predictions = box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND]
box_encodings = tf.concat(box_predictions[box_predictor.BOX_ENCODINGS],
axis=1)
objectness_predictions = tf.concat(box_predictions[
box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND], axis=1)
init_op = tf.global_variables_initializer()
resolution = 32
......
research/object_detection/core/losses.py
View file @ 78d5f8f8
......@@ -116,14 +116,23 @@ class WeightedL2LocalizationLoss(Loss):
class WeightedSmoothL1LocalizationLoss(Loss):
"""Smooth L1 localization loss function.
"""Smooth L1 localization loss function aka Huber Loss..
The smooth L1_loss is defined elementwise as .5 x^2 if |x|<1 and |x|-.5
otherwise, where x is the difference between predictions and target.
The smooth L1_loss is defined elementwise as .5 x^2 if |x| <= delta and
0.5 x^2 + delta * (|x|-delta) otherwise, where x is the difference between
predictions and target.
See also Equation (3) in the Fast R-CNN paper by Ross Girshick (ICCV 2015)
"""
def __init__(self, delta=1.0):
"""Constructor.
Args:
delta: delta for smooth L1 loss.
"""
self._delta = delta
def _compute_loss(self, prediction_tensor, target_tensor, weights):
"""Compute loss function.
......@@ -138,13 +147,14 @@ class WeightedSmoothL1LocalizationLoss(Loss):
loss: a float tensor of shape [batch_size, num_anchors] tensor
representing the value of the loss function.
"""
diff = prediction_tensor - target_tensor
abs_diff = tf.abs(diff)
abs_diff_lt_1 = tf.less(abs_diff, 1)
anchorwise_smooth_l1norm = tf.reduce_sum(
tf.where(abs_diff_lt_1, 0.5 * tf.square(abs_diff), abs_diff - 0.5),
2) * weights
return anchorwise_smooth_l1norm
return tf.reduce_sum(tf.losses.huber_loss(
target_tensor,
prediction_tensor,
delta=self._delta,
weights=tf.expand_dims(weights, axis=2),
loss_collection=None,
reduction=tf.losses.Reduction.NONE
), axis=2)
class WeightedIOULocalizationLoss(Loss):
......
research/object_detection/core/losses_test.py
View file @ 78d5f8f8
......@@ -545,7 +545,7 @@ class WeightedSoftmaxClassificationLossTest(tf.test.TestCase):
def testReturnsCorrectAnchorWiseLossWithHighLogitScaleSetting(self):
"""At very high logit_scale, all predictions will be ~0.33."""
# TODO: Also test logit_scale with anchorwise=False.
# TODO(yonib): Also test logit_scale with anchorwise=False.
logit_scale = 10e16
prediction_tensor = tf.constant([[[-100, 100, -100],
[100, -100, -100],
......
research/object_detection/core/preprocessor.py
View file @ 78d5f8f8
......@@ -204,7 +204,7 @@ def _random_integer(minval, maxval, seed):
[], minval=minval, maxval=maxval, dtype=tf.int32, seed=seed)
# TODO: This method is needed because the current
# TODO(mttang): This method is needed because the current
# tf.image.rgb_to_grayscale method does not support quantization. Replace with
# tf.image.rgb_to_grayscale after quantization support is added.
def _rgb_to_grayscale(images, name=None):
......@@ -2140,7 +2140,7 @@ def resize_to_range(image,
return result
# TODO: Make sure the static shapes are preserved.
# TODO(alirezafathi): Make sure the static shapes are preserved.
def resize_to_min_dimension(image, masks=None, min_dimension=600):
"""Resizes image and masks given the min size maintaining the aspect ratio.
......@@ -2226,7 +2226,7 @@ def scale_boxes_to_pixel_coordinates(image, boxes, keypoints=None):
return tuple(result)
# TODO: Investigate if instead the function should return None if
# TODO(alirezafathi): Investigate if instead the function should return None if
# masks is None.
# pylint: disable=g-doc-return-or-yield
def resize_image(image,
......
research/object_detection/core/target_assigner.py
View file @ 78d5f8f8
......@@ -324,7 +324,7 @@ class TargetAssigner(object):
return self._box_coder
# TODO: This method pulls in all the implementation dependencies into
# TODO(rathodv): This method pulls in all the implementation dependencies into
# core. Therefore its best to have this factory method outside of core.
def create_target_assigner(reference, stage=None,
negative_class_weight=1.0,
......
research/object_detection/core/target_assigner_test.py
View file @ 78d5f8f8
......@@ -836,7 +836,7 @@ class CreateTargetAssignerTest(tf.test.TestCase):
def test_create_target_assigner(self):
"""Tests that named constructor gives working target assigners.
TODO: Make this test more general.
TODO(rathodv): Make this test more general.
"""
corners = [[0.0, 0.0, 1.0, 1.0]]
groundtruth = box_list.BoxList(tf.constant(corners))
......
research/object_detection/data/BUILD deleted 100644 → 0
View file @ 629adffa
package(
default_visibility = ["//visibility:public"],
)
licenses(["notice"])
exports_files([
"pet_label_map.pbtxt",
])
research/object_detection/data_decoders/BUILD deleted 100644 → 0
View file @ 629adffa
# Tensorflow Object Detection API: data decoders.
package(
default_visibility = ["//visibility:public"],
)
licenses(["notice"])
# Apache 2.0
py_library(
name = "tf_example_decoder",
srcs = ["tf_example_decoder.py"],
deps = [
"//tensorflow",
"//tensorflow/models/research/object_detection/core:data_decoder",
"//tensorflow/models/research/object_detection/core:standard_fields",
"//tensorflow/models/research/object_detection/protos:input_reader_py_pb2",
"//tensorflow/models/research/object_detection/utils:label_map_util",
],
)
py_test(
name = "tf_example_decoder_test",
srcs = ["tf_example_decoder_test.py"],
deps = [
":tf_example_decoder",
"//tensorflow",
"//tensorflow/models/research/object_detection/core:standard_fields",
"//tensorflow/models/research/object_detection/protos:input_reader_py_pb2",
],
)
research/object_detection/data_decoders/tf_example_decoder.py
View file @ 78d5f8f8
......@@ -101,13 +101,18 @@ class TfExampleDecoder(data_decoder.DataDecoder):
'image/object/weight':
tf.VarLenFeature(tf.float32),
}
if dct_method:
image = slim_example_decoder.Image(
image_key='image/encoded',
format_key='image/format',
channels=3,
dct_method=dct_method)
else:
image = slim_example_decoder.Image(
image_key='image/encoded', format_key='image/format', channels=3)
self.items_to_handlers = {
fields.InputDataFields.image:
slim_example_decoder.Image(
image_key='image/encoded',
format_key='image/format',
channels=3,
dct_method=dct_method),
image,
fields.InputDataFields.source_id: (
slim_example_decoder.Tensor('image/source_id')),
fields.InputDataFields.key: (
......
research/object_detection/dataset_tools/BUILD deleted 100644 → 0
View file @ 629adffa
# Tensorflow Object Detection API: dataset tools.
package(
default_visibility = ["//visibility:public"],
)
licenses(["notice"])
# Apache 2.0
py_binary(
name = "create_coco_tf_record",
srcs = [
"create_coco_tf_record.py",
],
deps = [
"//PIL:pil",
"//pycocotools",
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:dataset_util",
"//tensorflow/models/research/object_detection/utils:label_map_util",
],
)
py_test(
name = "create_coco_tf_record_test",
srcs = [
"create_coco_tf_record_test.py",
],
deps = [
":create_coco_tf_record",
"//tensorflow",
],
)
py_binary(
name = "create_kitti_tf_record",
srcs = [
"create_kitti_tf_record.py",
],
deps = [
"//PIL:pil",
"//lxml",
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:dataset_util",
"//tensorflow/models/research/object_detection/utils:label_map_util",
"//tensorflow/models/research/object_detection/utils:np_box_ops",
],
)
py_test(
name = "create_kitti_tf_record_test",
srcs = [
"create_kitti_tf_record_test.py",
],
deps = [
":create_kitti_tf_record",
"//tensorflow",
],
)
py_binary(
name = "create_pascal_tf_record",
srcs = [
"create_pascal_tf_record.py",
],
deps = [
"//PIL:pil",
"//lxml",
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:dataset_util",
"//tensorflow/models/research/object_detection/utils:label_map_util",
],
)
py_test(
name = "create_pascal_tf_record_test",
srcs = [
"create_pascal_tf_record_test.py",
],
deps = [
":create_pascal_tf_record",
"//tensorflow",
],
)
py_binary(
name = "create_pet_tf_record",
srcs = [
"create_pet_tf_record.py",
],
deps = [
"//PIL:pil",
"//lxml",
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:dataset_util",
"//tensorflow/models/research/object_detection/utils:label_map_util",
],
)
py_library(
name = "oid_tfrecord_creation",
srcs = ["oid_tfrecord_creation.py"],
deps = [
"//tensorflow",
"//tensorflow/models/research/object_detection/core:standard_fields",
"//tensorflow/models/research/object_detection/utils:dataset_util",
],
)
py_test(
name = "oid_tfrecord_creation_test",
srcs = ["oid_tfrecord_creation_test.py"],
deps = [
":oid_tfrecord_creation",
"//contextlib2",
"//pandas",
"//tensorflow",
],
)
py_binary(
name = "create_oid_tf_record",
srcs = ["create_oid_tf_record.py"],
deps = [
":oid_tfrecord_creation",
"//contextlib2",
"//pandas",
"//tensorflow",
"//tensorflow/models/research/object_detection/utils:label_map_util",
],
)
research/object_detection/dataset_tools/create_kitti_tf_record.py
View file @ 78d5f8f8
......@@ -120,7 +120,7 @@ def convert_kitti_to_tfrecords(data_dir, output_path, classes_to_use,
# Filter all bounding boxes of this frame that are of a legal class, and
# don't overlap with a dontcare region.
# TODO filter out targets that are truncated or heavily occluded.
# TODO(talremez) filter out targets that are truncated or heavily occluded.
annotation_for_image = filter_annotations(img_anno, classes_to_use)
example = prepare_example(image_path, annotation_for_image, label_map_dict)
......
research/object_detection/dataset_tools/create_pet_tf_record.py
View file @ 78d5f8f8
......@@ -256,7 +256,7 @@ def create_tf_record(output_filename,
writer.close()
# TODO: Add test for pet/PASCAL main files.
# TODO(derekjchow): Add test for pet/PASCAL main files.
def main(_):
data_dir = FLAGS.data_dir
label_map_dict = label_map_util.get_label_map_dict(FLAGS.label_map_path)
......
research/object_detection/eval_util.py
View file @ 78d5f8f8
......@@ -50,7 +50,7 @@ def write_metrics(metrics, global_step, summary_dir):
logging.info('Metrics written to tf summary.')
# TODO: Add tests.
# TODO(rathodv): Add tests.
def visualize_detection_results(result_dict,
tag,
global_step,
......@@ -289,7 +289,7 @@ def _run_checkpoint_once(tensor_dict,
for evaluator in evaluators:
# TODO(b/65130867): Use image_id tensor once we fix the input data
# decoders to return correct image_id.
# TODO: result_dict contains batches of images, while
# TODO(akuznetsa): result_dict contains batches of images, while
# add_single_ground_truth_image_info expects a single image. Fix
evaluator.add_single_ground_truth_image_info(
image_id=batch, groundtruth_dict=result_dict)
......@@ -314,7 +314,7 @@ def _run_checkpoint_once(tensor_dict,
return (global_step, all_evaluator_metrics)
# TODO: Add tests.
# TODO(rathodv): Add tests.
def repeated_checkpoint_run(tensor_dict,
summary_dir,
evaluators,
......@@ -507,7 +507,7 @@ def result_dict_for_single_example(image,
if detection_fields.detection_masks in detections:
detection_masks = detections[detection_fields.detection_masks][0]
# TODO: This should be done in model's postprocess
# TODO(rathodv): This should be done in model's postprocess
# function ideally.
num_detections = tf.to_int32(detections[detection_fields.num_detections][0])
detection_boxes = tf.slice(
......
research/object_detection/evaluator.py
View file @ 78d5f8f8
......@@ -128,7 +128,7 @@ def get_evaluators(eval_config, categories):
def evaluate(create_input_dict_fn, create_model_fn, eval_config, categories,
checkpoint_dir, eval_dir, graph_hook_fn=None):
checkpoint_dir, eval_dir, graph_hook_fn=None, evaluator_list=None):
"""Evaluation function for detection models.
Args:
......@@ -143,6 +143,8 @@ def evaluate(create_input_dict_fn, create_model_fn, eval_config, categories,
completely built. This is helpful to perform additional changes to the
training graph such as optimizing batchnorm. The function should modify
the default graph.
evaluator_list: Optional list of instances of DetectionEvaluator. If not
given, this list of metrics is created according to the eval_config.
Returns:
metrics: A dictionary containing metric names and values from the latest
......@@ -222,10 +224,13 @@ def evaluate(create_input_dict_fn, create_model_fn, eval_config, categories,
latest_checkpoint = tf.train.latest_checkpoint(checkpoint_dir)
saver.restore(sess, latest_checkpoint)
if not evaluator_list:
evaluator_list = get_evaluators(eval_config, categories)
metrics = eval_util.repeated_checkpoint_run(
tensor_dict=tensor_dict,
summary_dir=eval_dir,
evaluators=get_evaluators(eval_config, categories),
evaluators=evaluator_list,
batch_processor=_process_batch,
checkpoint_dirs=[checkpoint_dir],
variables_to_restore=None,
......
research/object_detection/exporter.py
View file @ 78d5f8f8
......@@ -33,7 +33,7 @@ from object_detection.data_decoders import tf_example_decoder
slim = tf.contrib.slim
# TODO: Replace with freeze_graph.freeze_graph_with_def_protos when
# TODO(derekjchow): 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,
......@@ -242,7 +242,7 @@ def _add_output_tensor_nodes(postprocessed_tensors,
return outputs
def _write_frozen_graph(frozen_graph_path, frozen_graph_def):
def write_frozen_graph(frozen_graph_path, frozen_graph_def):
"""Writes frozen graph to disk.
Args:
......@@ -254,10 +254,10 @@ def _write_frozen_graph(frozen_graph_path, frozen_graph_def):
logging.info('%d ops in the final graph.', len(frozen_graph_def.node))
def _write_saved_model(saved_model_path,
frozen_graph_def,
inputs,
outputs):
def write_saved_model(saved_model_path,
frozen_graph_def,
inputs,
outputs):
"""Writes SavedModel to disk.
If checkpoint_path is not None bakes the weights into the graph thereby
......@@ -301,10 +301,11 @@ def _write_saved_model(saved_model_path,
builder.save()
def _write_graph_and_checkpoint(inference_graph_def,
model_path,
input_saver_def,
trained_checkpoint_prefix):
def write_graph_and_checkpoint(inference_graph_def,
model_path,
input_saver_def,
trained_checkpoint_prefix):
"""Writes the graph and the checkpoint into disk."""
for node in inference_graph_def.node:
node.device = ''
with tf.Graph().as_default():
......@@ -316,6 +317,44 @@ def _write_graph_and_checkpoint(inference_graph_def,
saver.save(sess, model_path)
def _get_outputs_from_inputs(input_tensors, detection_model,
output_collection_name):
inputs = tf.to_float(input_tensors)
preprocessed_inputs, true_image_shapes = detection_model.preprocess(inputs)
output_tensors = detection_model.predict(
preprocessed_inputs, true_image_shapes)
postprocessed_tensors = detection_model.postprocess(
output_tensors, true_image_shapes)
return _add_output_tensor_nodes(postprocessed_tensors,
output_collection_name)
def _build_detection_graph(input_type, detection_model, input_shape,
output_collection_name, graph_hook_fn):
"""Build the detection graph."""
if input_type not in input_placeholder_fn_map:
raise ValueError('Unknown input type: {}'.format(input_type))
placeholder_args = {}
if input_shape is not None:
if input_type != 'image_tensor':
raise ValueError('Can only specify input shape for `image_tensor` '
'inputs.')
placeholder_args['input_shape'] = input_shape
placeholder_tensor, input_tensors = input_placeholder_fn_map[input_type](
**placeholder_args)
outputs = _get_outputs_from_inputs(
input_tensors=input_tensors,
detection_model=detection_model,
output_collection_name=output_collection_name)
# Add global step to the graph.
slim.get_or_create_global_step()
if graph_hook_fn: graph_hook_fn()
return outputs, placeholder_tensor
def _export_inference_graph(input_type,
detection_model,
use_moving_averages,
......@@ -332,28 +371,12 @@ def _export_inference_graph(input_type,
saved_model_path = os.path.join(output_directory, 'saved_model')
model_path = os.path.join(output_directory, 'model.ckpt')
if input_type not in input_placeholder_fn_map:
raise ValueError('Unknown input type: {}'.format(input_type))
placeholder_args = {}
if input_shape is not None:
if input_type != 'image_tensor':
raise ValueError('Can only specify input shape for `image_tensor` '
'inputs.')
placeholder_args['input_shape'] = input_shape
placeholder_tensor, input_tensors = input_placeholder_fn_map[input_type](
**placeholder_args)
inputs = tf.to_float(input_tensors)
preprocessed_inputs, true_image_shapes = detection_model.preprocess(inputs)
output_tensors = detection_model.predict(
preprocessed_inputs, true_image_shapes)
postprocessed_tensors = detection_model.postprocess(
output_tensors, true_image_shapes)
outputs = _add_output_tensor_nodes(postprocessed_tensors,
output_collection_name)
# Add global step to the graph.
slim.get_or_create_global_step()
if graph_hook_fn: graph_hook_fn()
outputs, placeholder_tensor = _build_detection_graph(
input_type=input_type,
detection_model=detection_model,
input_shape=input_shape,
output_collection_name=output_collection_name,
graph_hook_fn=graph_hook_fn)
saver_kwargs = {}
if use_moving_averages:
......@@ -373,7 +396,7 @@ def _export_inference_graph(input_type,
saver = tf.train.Saver(**saver_kwargs)
input_saver_def = saver.as_saver_def()
_write_graph_and_checkpoint(
write_graph_and_checkpoint(
inference_graph_def=tf.get_default_graph().as_graph_def(),
model_path=model_path,
input_saver_def=input_saver_def,
......@@ -393,9 +416,9 @@ def _export_inference_graph(input_type,
filename_tensor_name='save/Const:0',
clear_devices=True,
initializer_nodes='')
_write_frozen_graph(frozen_graph_path, frozen_graph_def)
_write_saved_model(saved_model_path, frozen_graph_def,
placeholder_tensor, outputs)
write_frozen_graph(frozen_graph_path, frozen_graph_def)
write_saved_model(saved_model_path, frozen_graph_def,
placeholder_tensor, outputs)
def export_inference_graph(input_type,
......
research/object_detection/exporter_test.py
View file @ 78d5f8f8
......@@ -497,6 +497,66 @@ class ExportInferenceGraphTest(tf.test.TestCase):
self.assertAllClose(masks_np, np.arange(64).reshape([2, 2, 4, 4]))
self.assertAllClose(num_detections_np, [2, 1])
def test_write_frozen_graph(self):
tmp_dir = self.get_temp_dir()
trained_checkpoint_prefix = os.path.join(tmp_dir, 'model.ckpt')
self._save_checkpoint_from_mock_model(trained_checkpoint_prefix,
use_moving_averages=True)
output_directory = os.path.join(tmp_dir, 'output')
inference_graph_path = os.path.join(output_directory,
'frozen_inference_graph.pb')
tf.gfile.MakeDirs(output_directory)
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
detection_model = model_builder.build(pipeline_config.model,
is_training=False)
outputs, _ = exporter._build_detection_graph(
input_type='tf_example',
detection_model=detection_model,
input_shape=None,
output_collection_name='inference_op',
graph_hook_fn=None)
output_node_names = ','.join(outputs.keys())
saver = tf.train.Saver()
input_saver_def = saver.as_saver_def()
frozen_graph_def = exporter.freeze_graph_with_def_protos(
input_graph_def=tf.get_default_graph().as_graph_def(),
input_saver_def=input_saver_def,
input_checkpoint=trained_checkpoint_prefix,
output_node_names=output_node_names,
restore_op_name='save/restore_all',
filename_tensor_name='save/Const:0',
clear_devices=True,
initializer_nodes='')
exporter.write_frozen_graph(inference_graph_path, frozen_graph_def)
inference_graph = self._load_inference_graph(inference_graph_path)
tf_example_np = np.expand_dims(self._create_tf_example(
np.ones((4, 4, 3)).astype(np.uint8)), axis=0)
with self.test_session(graph=inference_graph) as sess:
tf_example = inference_graph.get_tensor_by_name('tf_example: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')
(boxes_np, scores_np, classes_np, masks_np, num_detections_np) = sess.run(
[boxes, scores, classes, masks, num_detections],
feed_dict={tf_example: tf_example_np})
self.assertAllClose(boxes_np, [[[0.0, 0.0, 0.5, 0.5],
[0.5, 0.5, 0.8, 0.8]],
[[0.5, 0.5, 1.0, 1.0],
[0.0, 0.0, 0.0, 0.0]]])
self.assertAllClose(scores_np, [[0.7, 0.6],
[0.9, 0.0]])
self.assertAllClose(classes_np, [[1, 2],
[2, 1]])
self.assertAllClose(masks_np, np.arange(64).reshape([2, 2, 4, 4]))
self.assertAllClose(num_detections_np, [2, 1])
def test_export_graph_saves_pipeline_file(self):
tmp_dir = self.get_temp_dir()
trained_checkpoint_prefix = os.path.join(tmp_dir, 'model.ckpt')
......@@ -578,6 +638,82 @@ class ExportInferenceGraphTest(tf.test.TestCase):
self.assertAllClose(masks_np, np.arange(64).reshape([2, 2, 4, 4]))
self.assertAllClose(num_detections_np, [2, 1])
def test_write_saved_model(self):
tmp_dir = self.get_temp_dir()
trained_checkpoint_prefix = os.path.join(tmp_dir, 'model.ckpt')
self._save_checkpoint_from_mock_model(trained_checkpoint_prefix,
use_moving_averages=False)
output_directory = os.path.join(tmp_dir, 'output')
saved_model_path = os.path.join(output_directory, 'saved_model')
tf.gfile.MakeDirs(output_directory)
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
detection_model = model_builder.build(pipeline_config.model,
is_training=False)
outputs, placeholder_tensor = exporter._build_detection_graph(
input_type='tf_example',
detection_model=detection_model,
input_shape=None,
output_collection_name='inference_op',
graph_hook_fn=None)
output_node_names = ','.join(outputs.keys())
saver = tf.train.Saver()
input_saver_def = saver.as_saver_def()
frozen_graph_def = exporter.freeze_graph_with_def_protos(
input_graph_def=tf.get_default_graph().as_graph_def(),
input_saver_def=input_saver_def,
input_checkpoint=trained_checkpoint_prefix,
output_node_names=output_node_names,
restore_op_name='save/restore_all',
filename_tensor_name='save/Const:0',
clear_devices=True,
initializer_nodes='')
exporter.write_saved_model(
saved_model_path=saved_model_path,
frozen_graph_def=frozen_graph_def,
inputs=placeholder_tensor,
outputs=outputs)
tf_example_np = np.hstack([self._create_tf_example(
np.ones((4, 4, 3)).astype(np.uint8))] * 2)
with tf.Graph().as_default() as od_graph:
with self.test_session(graph=od_graph) as sess:
meta_graph = tf.saved_model.loader.load(
sess, [tf.saved_model.tag_constants.SERVING], saved_model_path)
signature = meta_graph.signature_def['serving_default']
input_tensor_name = signature.inputs['inputs'].name
tf_example = od_graph.get_tensor_by_name(input_tensor_name)
boxes = od_graph.get_tensor_by_name(
signature.outputs['detection_boxes'].name)
scores = od_graph.get_tensor_by_name(
signature.outputs['detection_scores'].name)
classes = od_graph.get_tensor_by_name(
signature.outputs['detection_classes'].name)
masks = od_graph.get_tensor_by_name(
signature.outputs['detection_masks'].name)
num_detections = od_graph.get_tensor_by_name(
signature.outputs['num_detections'].name)
(boxes_np, scores_np, classes_np, masks_np,
num_detections_np) = sess.run(
[boxes, scores, classes, masks, num_detections],
feed_dict={tf_example: tf_example_np})
self.assertAllClose(boxes_np, [[[0.0, 0.0, 0.5, 0.5],
[0.5, 0.5, 0.8, 0.8]],
[[0.5, 0.5, 1.0, 1.0],
[0.0, 0.0, 0.0, 0.0]]])
self.assertAllClose(scores_np, [[0.7, 0.6],
[0.9, 0.0]])
self.assertAllClose(classes_np, [[1, 2],
[2, 1]])
self.assertAllClose(masks_np, np.arange(64).reshape([2, 2, 4, 4]))
self.assertAllClose(num_detections_np, [2, 1])
def test_export_checkpoint_and_run_inference(self):
tmp_dir = self.get_temp_dir()
trained_checkpoint_prefix = os.path.join(tmp_dir, 'model.ckpt')
......@@ -626,6 +762,64 @@ class ExportInferenceGraphTest(tf.test.TestCase):
self.assertAllClose(masks_np, np.arange(64).reshape([2, 2, 4, 4]))
self.assertAllClose(num_detections_np, [2, 1])
def test_write_graph_and_checkpoint(self):
tmp_dir = self.get_temp_dir()
trained_checkpoint_prefix = os.path.join(tmp_dir, 'model.ckpt')
self._save_checkpoint_from_mock_model(trained_checkpoint_prefix,
use_moving_averages=False)
output_directory = os.path.join(tmp_dir, 'output')
model_path = os.path.join(output_directory, 'model.ckpt')
meta_graph_path = model_path + '.meta'
tf.gfile.MakeDirs(output_directory)
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
detection_model = model_builder.build(pipeline_config.model,
is_training=False)
exporter._build_detection_graph(
input_type='tf_example',
detection_model=detection_model,
input_shape=None,
output_collection_name='inference_op',
graph_hook_fn=None)
saver = tf.train.Saver()
input_saver_def = saver.as_saver_def()
exporter.write_graph_and_checkpoint(
inference_graph_def=tf.get_default_graph().as_graph_def(),
model_path=model_path,
input_saver_def=input_saver_def,
trained_checkpoint_prefix=trained_checkpoint_prefix)
tf_example_np = np.hstack([self._create_tf_example(
np.ones((4, 4, 3)).astype(np.uint8))] * 2)
with tf.Graph().as_default() as od_graph:
with self.test_session(graph=od_graph) as sess:
new_saver = tf.train.import_meta_graph(meta_graph_path)
new_saver.restore(sess, model_path)
tf_example = od_graph.get_tensor_by_name('tf_example:0')
boxes = od_graph.get_tensor_by_name('detection_boxes:0')
scores = od_graph.get_tensor_by_name('detection_scores:0')
classes = od_graph.get_tensor_by_name('detection_classes:0')
masks = od_graph.get_tensor_by_name('detection_masks:0')
num_detections = od_graph.get_tensor_by_name('num_detections:0')
(boxes_np, scores_np, classes_np, masks_np,
num_detections_np) = sess.run(
[boxes, scores, classes, masks, num_detections],
feed_dict={tf_example: tf_example_np})
self.assertAllClose(boxes_np, [[[0.0, 0.0, 0.5, 0.5],
[0.5, 0.5, 0.8, 0.8]],
[[0.5, 0.5, 1.0, 1.0],
[0.0, 0.0, 0.0, 0.0]]])
self.assertAllClose(scores_np, [[0.7, 0.6],
[0.9, 0.0]])
self.assertAllClose(classes_np, [[1, 2],
[2, 1]])
self.assertAllClose(masks_np, np.arange(64).reshape([2, 2, 4, 4]))
self.assertAllClose(num_detections_np, [2, 1])
if __name__ == '__main__':
tf.test.main()
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