Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
ModelZoo
ResNet50_tensorflow
Commits
582bf927
Commit
582bf927
authored
Jul 27, 2017
by
derekjchow
Committed by
GitHub
Jul 27, 2017
Browse files
Merge pull request #2053 from derekjchow/master
object_detection exporter updates
parents
ecf5edf1
a2cb67c2
Changes
23
Expand all
Hide whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
1299 additions
and
828 deletions
+1299
-828
object_detection/core/BUILD
object_detection/core/BUILD
+1
-0
object_detection/core/box_predictor.py
object_detection/core/box_predictor.py
+18
-17
object_detection/core/model.py
object_detection/core/model.py
+7
-8
object_detection/core/post_processing.py
object_detection/core/post_processing.py
+71
-57
object_detection/core/post_processing_test.py
object_detection/core/post_processing_test.py
+180
-68
object_detection/core/preprocessor.py
object_detection/core/preprocessor.py
+86
-52
object_detection/core/preprocessor_test.py
object_detection/core/preprocessor_test.py
+31
-45
object_detection/export_inference_graph.py
object_detection/export_inference_graph.py
+38
-33
object_detection/exporter.py
object_detection/exporter.py
+183
-146
object_detection/exporter_test.py
object_detection/exporter_test.py
+241
-128
object_detection/meta_architectures/BUILD
object_detection/meta_architectures/BUILD
+2
-3
object_detection/meta_architectures/faster_rcnn_meta_arch.py
object_detection/meta_architectures/faster_rcnn_meta_arch.py
+180
-96
object_detection/meta_architectures/faster_rcnn_meta_arch_test_lib.py
...tion/meta_architectures/faster_rcnn_meta_arch_test_lib.py
+124
-89
object_detection/meta_architectures/ssd_meta_arch.py
object_detection/meta_architectures/ssd_meta_arch.py
+47
-27
object_detection/meta_architectures/ssd_meta_arch_test.py
object_detection/meta_architectures/ssd_meta_arch_test.py
+68
-33
object_detection/models/BUILD
object_detection/models/BUILD
+0
-1
object_detection/models/faster_rcnn_inception_resnet_v2_feature_extractor.py
...dels/faster_rcnn_inception_resnet_v2_feature_extractor.py
+9
-16
object_detection/trainer.py
object_detection/trainer.py
+8
-2
object_detection/trainer_test.py
object_detection/trainer_test.py
+4
-7
object_detection/utils/BUILD
object_detection/utils/BUILD
+1
-0
No files found.
object_detection/core/BUILD
View file @
582bf927
...
@@ -270,6 +270,7 @@ py_library(
...
@@ -270,6 +270,7 @@ py_library(
deps
=
[
deps
=
[
"//tensorflow"
,
"//tensorflow"
,
"//tensorflow_models/object_detection/utils:ops"
,
"//tensorflow_models/object_detection/utils:ops"
,
"//tensorflow_models/object_detection/utils:shape_utils"
,
"//tensorflow_models/object_detection/utils:static_shape"
,
"//tensorflow_models/object_detection/utils:static_shape"
,
],
],
)
)
...
...
object_detection/core/box_predictor.py
View file @
582bf927
...
@@ -29,6 +29,7 @@ few box predictor architectures are shared across many models.
...
@@ -29,6 +29,7 @@ few box predictor architectures are shared across many models.
from
abc
import
abstractmethod
from
abc
import
abstractmethod
import
tensorflow
as
tf
import
tensorflow
as
tf
from
object_detection.utils
import
ops
from
object_detection.utils
import
ops
from
object_detection.utils
import
shape_utils
from
object_detection.utils
import
static_shape
from
object_detection.utils
import
static_shape
slim
=
tf
.
contrib
.
slim
slim
=
tf
.
contrib
.
slim
...
@@ -316,6 +317,8 @@ class MaskRCNNBoxPredictor(BoxPredictor):
...
@@ -316,6 +317,8 @@ class MaskRCNNBoxPredictor(BoxPredictor):
self
.
_predict_instance_masks
=
predict_instance_masks
self
.
_predict_instance_masks
=
predict_instance_masks
self
.
_mask_prediction_conv_depth
=
mask_prediction_conv_depth
self
.
_mask_prediction_conv_depth
=
mask_prediction_conv_depth
self
.
_predict_keypoints
=
predict_keypoints
self
.
_predict_keypoints
=
predict_keypoints
if
self
.
_predict_instance_masks
:
raise
ValueError
(
'Mask prediction is unimplemented.'
)
if
self
.
_predict_keypoints
:
if
self
.
_predict_keypoints
:
raise
ValueError
(
'Keypoint prediction is unimplemented.'
)
raise
ValueError
(
'Keypoint prediction is unimplemented.'
)
if
((
self
.
_predict_instance_masks
or
self
.
_predict_keypoints
)
and
if
((
self
.
_predict_instance_masks
or
self
.
_predict_keypoints
)
and
...
@@ -524,23 +527,21 @@ class ConvolutionalBoxPredictor(BoxPredictor):
...
@@ -524,23 +527,21 @@ class ConvolutionalBoxPredictor(BoxPredictor):
class_predictions_with_background
=
tf
.
sigmoid
(
class_predictions_with_background
=
tf
.
sigmoid
(
class_predictions_with_background
)
class_predictions_with_background
)
batch_size
=
static_shape
.
get_batch_size
(
image_features
.
get_shape
())
combined_feature_map_shape
=
shape_utils
.
combined_static_and_dynamic_shape
(
if
batch_size
is
None
:
image_features
)
features_height
=
static_shape
.
get_height
(
image_features
.
get_shape
())
box_encodings
=
tf
.
reshape
(
features_width
=
static_shape
.
get_width
(
image_features
.
get_shape
())
box_encodings
,
tf
.
stack
([
combined_feature_map_shape
[
0
],
flattened_predictions_size
=
(
features_height
*
features_width
*
combined_feature_map_shape
[
1
]
*
num_predictions_per_location
)
combined_feature_map_shape
[
2
]
*
box_encodings
=
tf
.
reshape
(
num_predictions_per_location
,
box_encodings
,
1
,
self
.
_box_code_size
]))
[
-
1
,
flattened_predictions_size
,
1
,
self
.
_box_code_size
])
class_predictions_with_background
=
tf
.
reshape
(
class_predictions_with_background
=
tf
.
reshape
(
class_predictions_with_background
,
class_predictions_with_background
,
tf
.
stack
([
combined_feature_map_shape
[
0
],
[
-
1
,
flattened_predictions_size
,
num_class_slots
])
combined_feature_map_shape
[
1
]
*
else
:
combined_feature_map_shape
[
2
]
*
box_encodings
=
tf
.
reshape
(
num_predictions_per_location
,
box_encodings
,
[
batch_size
,
-
1
,
1
,
self
.
_box_code_size
])
num_class_slots
]))
class_predictions_with_background
=
tf
.
reshape
(
class_predictions_with_background
,
[
batch_size
,
-
1
,
num_class_slots
])
return
{
BOX_ENCODINGS
:
box_encodings
,
return
{
BOX_ENCODINGS
:
box_encodings
,
CLASS_PREDICTIONS_WITH_BACKGROUND
:
CLASS_PREDICTIONS_WITH_BACKGROUND
:
class_predictions_with_background
}
class_predictions_with_background
}
object_detection/core/model.py
View file @
582bf927
...
@@ -228,25 +228,24 @@ class DetectionModel(object):
...
@@ -228,25 +228,24 @@ class DetectionModel(object):
fields
.
BoxListFields
.
keypoints
]
=
groundtruth_keypoints_list
fields
.
BoxListFields
.
keypoints
]
=
groundtruth_keypoints_list
@
abstractmethod
@
abstractmethod
def
restore_
fn
(
self
,
checkpoint_path
,
from_detection_checkpoint
=
True
):
def
restore_
map
(
self
,
from_detection_checkpoint
=
True
):
"""Return
callable for loading a foreign checkpoint into tensorflow graph
.
"""Return
s a map of variables to load from a foreign checkpoint
.
Loads variables from a different tensorflow graph (typically feature
Returns a map of variable names to load from a checkpoint to variables in
extractor variables)
. This enables the model to initialize based on weights
the model graph
. This enables the model to initialize based on weights
from
from
another task. For example, the feature extractor variables from a
another task. For example, the feature extractor variables from a
classification model can be used to bootstrap training of an object
classification model can be used to bootstrap training of an object
detector. When loading from an object detection model, the checkpoint model
detector. When loading from an object detection model, the checkpoint model
should have the same parameters as this detection model with exception of
should have the same parameters as this detection model with exception of
the num_classes parameter.
the num_classes parameter.
Args:
Args:
checkpoint_path: path to checkpoint to restore.
from_detection_checkpoint: whether to restore from a full detection
from_detection_checkpoint: whether to restore from a full detection
checkpoint (with compatible variable names) or to restore from a
checkpoint (with compatible variable names) or to restore from a
classification checkpoint for initialization prior to training.
classification checkpoint for initialization prior to training.
Returns:
Returns:
a callable which takes a tf.Session as input and loads a checkpoint whe
n
A dict mapping variable names (to load from a checkpoint) to variables i
n
run
.
the model graph
.
"""
"""
pass
pass
object_detection/core/post_processing.py
View file @
582bf927
...
@@ -174,7 +174,8 @@ def batch_multiclass_non_max_suppression(boxes,
...
@@ -174,7 +174,8 @@ def batch_multiclass_non_max_suppression(boxes,
change_coordinate_frame
=
False
,
change_coordinate_frame
=
False
,
num_valid_boxes
=
None
,
num_valid_boxes
=
None
,
masks
=
None
,
masks
=
None
,
scope
=
None
):
scope
=
None
,
parallel_iterations
=
32
):
"""Multi-class version of non maximum suppression that operates on a batch.
"""Multi-class version of non maximum suppression that operates on a batch.
This op is similar to `multiclass_non_max_suppression` but operates on a batch
This op is similar to `multiclass_non_max_suppression` but operates on a batch
...
@@ -208,26 +209,28 @@ def batch_multiclass_non_max_suppression(boxes,
...
@@ -208,26 +209,28 @@ def batch_multiclass_non_max_suppression(boxes,
float32 tensor containing box masks. `q` can be either number of classes
float32 tensor containing box masks. `q` can be either number of classes
or 1 depending on whether a separate mask is predicted per class.
or 1 depending on whether a separate mask is predicted per class.
scope: tf scope name.
scope: tf scope name.
parallel_iterations: (optional) number of batch items to process in
parallel.
Returns:
Returns:
A dictionary containing the following entries:
'nmsed_boxes': A [batch_size, max_detections, 4] float32 tensor
'detection_boxes': A [batch_size, max_detections, 4] float32 tensor
containing the non-max suppressed boxes.
containing the non-max suppressed boxes.
'
detection
_scores': A [bath_size, max_detections] float32 tensor containing
'
nmsed
_scores': A [bat
c
h_size, max_detections] float32 tensor containing
the scores for the boxes.
the scores for the boxes.
'
detection
_classes': A [batch_size, max_detections] float32 tensor
'
nmsed
_classes': A [batch_size, max_detections] float32 tensor
containing the class for boxes.
containing the class for boxes.
'num_detections': A [batchsize] float32 tensor indicating the number of
'nmsed_masks': (optional) a
[batch_size, max_detections, mask_height, mask_width] float32 tensor
containing masks for each selected box. This is set to None if input
`masks` is None.
'num_detections': A [batch_size] int32 tensor indicating the number of
valid detections per batch item. Only the top num_detections[i] entries in
valid detections per batch item. Only the top num_detections[i] entries in
nms_boxes[i], nms_scores[i] and nms_class[i] are valid. the rest of the
nms_boxes[i], nms_scores[i] and nms_class[i] are valid. the rest of the
entries are zero paddings.
entries are zero paddings.
'detection_masks': (optional) a
[batch_size, max_detections, mask_height, mask_width] float32 tensor
containing masks for each selected box.
Raises:
Raises:
ValueError: if
iou_thresh is not in [0, 1] or if input boxlist does not have
ValueError: if
`q` in boxes.shape is not 1 or not equal to number of
a valid scores field
.
classes as inferred from scores.shape
.
"""
"""
q
=
boxes
.
shape
[
2
].
value
q
=
boxes
.
shape
[
2
].
value
num_classes
=
scores
.
shape
[
2
].
value
num_classes
=
scores
.
shape
[
2
].
value
...
@@ -235,36 +238,45 @@ def batch_multiclass_non_max_suppression(boxes,
...
@@ -235,36 +238,45 @@ def batch_multiclass_non_max_suppression(boxes,
raise
ValueError
(
'third dimension of boxes must be either 1 or equal '
raise
ValueError
(
'third dimension of boxes must be either 1 or equal '
'to the third dimension of scores'
)
'to the third dimension of scores'
)
original_masks
=
masks
with
tf
.
name_scope
(
scope
,
'BatchMultiClassNonMaxSuppression'
):
with
tf
.
name_scope
(
scope
,
'BatchMultiClassNonMaxSuppression'
):
per_image_boxes_list
=
tf
.
unstack
(
boxes
)
boxes_shape
=
boxes
.
shape
per_image_scores_list
=
tf
.
unstack
(
scores
)
batch_size
=
boxes_shape
[
0
].
value
num_valid_boxes_list
=
len
(
per_image_boxes_list
)
*
[
None
]
num_anchors
=
boxes_shape
[
1
].
value
per_image_masks_list
=
len
(
per_image_boxes_list
)
*
[
None
]
if
num_valid_boxes
is
not
None
:
if
batch_size
is
None
:
num_valid_boxes_list
=
tf
.
unstack
(
num_valid_boxes
)
batch_size
=
tf
.
shape
(
boxes
)[
0
]
if
masks
is
not
None
:
if
num_anchors
is
None
:
per_image_masks_list
=
tf
.
unstack
(
masks
)
num_anchors
=
tf
.
shape
(
boxes
)[
1
]
# If num valid boxes aren't provided, create one and mark all boxes as
# valid.
if
num_valid_boxes
is
None
:
num_valid_boxes
=
tf
.
ones
([
batch_size
],
dtype
=
tf
.
int32
)
*
num_anchors
detection_boxes_list
=
[]
# If masks aren't provided, create dummy masks so we can only have one copy
detection_scores_list
=
[]
# of single_image_nms_fn and discard the dummy masks after map_fn.
detection_classes_list
=
[]
if
masks
is
None
:
num_detections_list
=
[]
masks_shape
=
tf
.
stack
([
batch_size
,
num_anchors
,
1
,
0
,
0
])
detection_masks_list
=
[]
masks
=
tf
.
zeros
(
masks_shape
)
for
(
per_image_boxes
,
per_image_scores
,
per_image_masks
,
num_valid_boxes
)
in
zip
(
per_image_boxes_list
,
per_image_scores_list
,
def
single_image_nms_fn
(
args
):
per_image_masks_list
,
num_valid_boxes_list
):
"""Runs NMS on a single image and returns padded output."""
if
num_valid_boxes
is
not
None
:
(
per_image_boxes
,
per_image_scores
,
per_image_masks
,
per_image_boxes
=
tf
.
reshape
(
per_image_num_valid_boxes
)
=
args
tf
.
slice
(
per_image_boxes
,
3
*
[
0
],
per_image_boxes
=
tf
.
reshape
(
tf
.
stack
([
num_valid_boxes
,
-
1
,
-
1
])),
[
-
1
,
q
,
4
])
tf
.
slice
(
per_image_boxes
,
3
*
[
0
],
per_image_scores
=
tf
.
reshape
(
tf
.
stack
([
per_image_num_valid_boxes
,
-
1
,
-
1
])),
[
-
1
,
q
,
4
])
tf
.
slice
(
per_image_scores
,
[
0
,
0
],
per_image_scores
=
tf
.
reshape
(
tf
.
stack
([
num_valid_boxes
,
-
1
])),
[
-
1
,
num_classes
])
tf
.
slice
(
per_image_scores
,
[
0
,
0
],
if
masks
is
not
None
:
tf
.
stack
([
per_image_num_valid_boxes
,
-
1
])),
per_image_masks
=
tf
.
reshape
(
[
-
1
,
num_classes
])
tf
.
slice
(
per_image_masks
,
4
*
[
0
],
tf
.
stack
([
num_valid_boxes
,
-
1
,
-
1
,
-
1
])),
per_image_masks
=
tf
.
reshape
(
[
-
1
,
q
,
masks
.
shape
[
3
].
value
,
masks
.
shape
[
4
].
value
])
tf
.
slice
(
per_image_masks
,
4
*
[
0
],
tf
.
stack
([
per_image_num_valid_boxes
,
-
1
,
-
1
,
-
1
])),
[
-
1
,
q
,
per_image_masks
.
shape
[
2
].
value
,
per_image_masks
.
shape
[
3
].
value
])
nmsed_boxlist
=
multiclass_non_max_suppression
(
nmsed_boxlist
=
multiclass_non_max_suppression
(
per_image_boxes
,
per_image_boxes
,
per_image_scores
,
per_image_scores
,
...
@@ -275,24 +287,26 @@ def batch_multiclass_non_max_suppression(boxes,
...
@@ -275,24 +287,26 @@ def batch_multiclass_non_max_suppression(boxes,
masks
=
per_image_masks
,
masks
=
per_image_masks
,
clip_window
=
clip_window
,
clip_window
=
clip_window
,
change_coordinate_frame
=
change_coordinate_frame
)
change_coordinate_frame
=
change_coordinate_frame
)
num_detections_list
.
append
(
tf
.
to_float
(
nmsed_boxlist
.
num_boxes
()))
padded_boxlist
=
box_list_ops
.
pad_or_clip_box_list
(
nmsed_boxlist
,
padded_boxlist
=
box_list_ops
.
pad_or_clip_box_list
(
nmsed_boxlist
,
max_total_size
)
max_total_size
)
detection_boxes_list
.
append
(
padded_boxlist
.
get
())
num_detections
=
nmsed_boxlist
.
num_boxes
()
detection_scores_list
.
append
(
nmsed_boxes
=
padded_boxlist
.
get
()
padded_boxlist
.
get_field
(
fields
.
BoxListFields
.
scores
))
nmsed_scores
=
padded_boxlist
.
get_field
(
fields
.
BoxListFields
.
scores
)
detection_classes_list
.
append
(
nmsed_classes
=
padded_boxlist
.
get_field
(
fields
.
BoxListFields
.
classes
)
padded_boxlist
.
get_field
(
fields
.
BoxListFields
.
classes
))
nmsed_masks
=
padded_boxlist
.
get_field
(
fields
.
BoxListFields
.
masks
)
if
masks
is
not
None
:
return
[
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
detection_masks_list
.
append
(
num_detections
]
padded_boxlist
.
get_field
(
fields
.
BoxListFields
.
masks
))
nms_dict
=
{
(
batch_nmsed_boxes
,
batch_nmsed_scores
,
'detection_boxes'
:
tf
.
stack
(
detection_boxes_list
),
batch_nmsed_classes
,
batch_nmsed_masks
,
'detection_scores'
:
tf
.
stack
(
detection_scores_list
),
batch_num_detections
)
=
tf
.
map_fn
(
'detection_classes'
:
tf
.
stack
(
detection_classes_list
),
single_image_nms_fn
,
'num_detections'
:
tf
.
stack
(
num_detections_list
)
elems
=
[
boxes
,
scores
,
masks
,
num_valid_boxes
],
}
dtype
=
[
tf
.
float32
,
tf
.
float32
,
tf
.
float32
,
tf
.
float32
,
tf
.
int32
],
if
masks
is
not
None
:
parallel_iterations
=
parallel_iterations
)
nms_dict
[
'detection_masks'
]
=
tf
.
stack
(
detection_masks_list
)
return
nms_dict
if
original_masks
is
None
:
batch_nmsed_masks
=
None
return
(
batch_nmsed_boxes
,
batch_nmsed_scores
,
batch_nmsed_classes
,
batch_nmsed_masks
,
batch_num_detections
)
object_detection/core/post_processing_test.py
View file @
582bf927
...
@@ -496,15 +496,21 @@ class MulticlassNonMaxSuppressionTest(tf.test.TestCase):
...
@@ -496,15 +496,21 @@ class MulticlassNonMaxSuppressionTest(tf.test.TestCase):
exp_nms_scores
=
[[.
95
,
.
9
,
.
85
,
.
3
]]
exp_nms_scores
=
[[.
95
,
.
9
,
.
85
,
.
3
]]
exp_nms_classes
=
[[
0
,
0
,
1
,
0
]]
exp_nms_classes
=
[[
0
,
0
,
1
,
0
]]
nms_dict
=
post_processing
.
batch_multiclass_non_max_suppression
(
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
boxes
,
scores
,
score_thresh
,
iou_thresh
,
num_detections
)
=
post_processing
.
batch_multiclass_non_max_suppression
(
max_size_per_class
=
max_output_size
,
max_total_size
=
max_output_size
)
boxes
,
scores
,
score_thresh
,
iou_thresh
,
max_size_per_class
=
max_output_size
,
max_total_size
=
max_output_size
)
self
.
assertIsNone
(
nmsed_masks
)
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
nms_output
=
sess
.
run
(
nms_dict
)
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
self
.
assertAllClose
(
nms_output
[
'detection_boxes'
],
exp_nms_corners
)
num_detections
)
=
sess
.
run
([
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
self
.
assertAllClose
(
nms_output
[
'detection_scores'
],
exp_nms_scores
)
num_detections
])
self
.
assertAllClose
(
nms_output
[
'detection_classes'
],
exp_nms_classes
)
self
.
assertAllClose
(
nmsed_boxes
,
exp_nms_corners
)
self
.
assertEqual
(
nms_output
[
'num_detections'
],
[
4
])
self
.
assertAllClose
(
nmsed_scores
,
exp_nms_scores
)
self
.
assertAllClose
(
nmsed_classes
,
exp_nms_classes
)
self
.
assertEqual
(
num_detections
,
[
4
])
def
test_batch_multiclass_nms_with_batch_size_2
(
self
):
def
test_batch_multiclass_nms_with_batch_size_2
(
self
):
boxes
=
tf
.
constant
([[[[
0
,
0
,
1
,
1
],
[
0
,
0
,
4
,
5
]],
boxes
=
tf
.
constant
([[[[
0
,
0
,
1
,
1
],
[
0
,
0
,
4
,
5
]],
...
@@ -524,28 +530,42 @@ class MulticlassNonMaxSuppressionTest(tf.test.TestCase):
...
@@ -524,28 +530,42 @@ class MulticlassNonMaxSuppressionTest(tf.test.TestCase):
iou_thresh
=
.
5
iou_thresh
=
.
5
max_output_size
=
4
max_output_size
=
4
exp_nms_corners
=
[[[
0
,
10
,
1
,
11
],
exp_nms_corners
=
np
.
array
([[[
0
,
10
,
1
,
11
],
[
0
,
0
,
1
,
1
],
[
0
,
0
,
1
,
1
],
[
0
,
0
,
0
,
0
],
[
0
,
0
,
0
,
0
],
[
0
,
0
,
0
,
0
]],
[
0
,
0
,
0
,
0
]],
[[
0
,
999
,
2
,
1004
],
[[
0
,
999
,
2
,
1004
],
[
0
,
10.1
,
1
,
11.1
],
[
0
,
10.1
,
1
,
11.1
],
[
0
,
100
,
1
,
101
],
[
0
,
100
,
1
,
101
],
[
0
,
0
,
0
,
0
]]]
[
0
,
0
,
0
,
0
]]])
exp_nms_scores
=
[[.
95
,
.
9
,
0
,
0
],
exp_nms_scores
=
np
.
array
([[.
95
,
.
9
,
0
,
0
],
[.
85
,
.
5
,
.
3
,
0
]]
[.
85
,
.
5
,
.
3
,
0
]])
exp_nms_classes
=
[[
0
,
0
,
0
,
0
],
exp_nms_classes
=
np
.
array
([[
0
,
0
,
0
,
0
],
[
1
,
0
,
0
,
0
]]
[
1
,
0
,
0
,
0
]])
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
num_detections
)
=
post_processing
.
batch_multiclass_non_max_suppression
(
boxes
,
scores
,
score_thresh
,
iou_thresh
,
max_size_per_class
=
max_output_size
,
max_total_size
=
max_output_size
)
self
.
assertIsNone
(
nmsed_masks
)
# Check static shapes
self
.
assertAllEqual
(
nmsed_boxes
.
shape
.
as_list
(),
exp_nms_corners
.
shape
)
self
.
assertAllEqual
(
nmsed_scores
.
shape
.
as_list
(),
exp_nms_scores
.
shape
)
self
.
assertAllEqual
(
nmsed_classes
.
shape
.
as_list
(),
exp_nms_classes
.
shape
)
self
.
assertEqual
(
num_detections
.
shape
.
as_list
(),
[
2
])
nms_dict
=
post_processing
.
batch_multiclass_non_max_suppression
(
boxes
,
scores
,
score_thresh
,
iou_thresh
,
max_size_per_class
=
max_output_size
,
max_total_size
=
max_output_size
)
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
nms_output
=
sess
.
run
(
nms_dict
)
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
self
.
assertAllClose
(
nms_output
[
'detection_boxes'
],
exp_nms_corners
)
num_detections
)
=
sess
.
run
([
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
self
.
assertAllClose
(
nms_output
[
'detection_scores'
],
exp_nms_scores
)
num_detections
])
self
.
assertAllClose
(
nms_output
[
'detection_classes'
],
exp_nms_classes
)
self
.
assertAllClose
(
nmsed_boxes
,
exp_nms_corners
)
self
.
assertAllClose
(
nms_output
[
'num_detections'
],
[
2
,
3
])
self
.
assertAllClose
(
nmsed_scores
,
exp_nms_scores
)
self
.
assertAllClose
(
nmsed_classes
,
exp_nms_classes
)
self
.
assertAllClose
(
num_detections
,
[
2
,
3
])
def
test_batch_multiclass_nms_with_masks
(
self
):
def
test_batch_multiclass_nms_with_masks
(
self
):
boxes
=
tf
.
constant
([[[[
0
,
0
,
1
,
1
],
[
0
,
0
,
4
,
5
]],
boxes
=
tf
.
constant
([[[[
0
,
0
,
1
,
1
],
[
0
,
0
,
4
,
5
]],
...
@@ -574,38 +594,126 @@ class MulticlassNonMaxSuppressionTest(tf.test.TestCase):
...
@@ -574,38 +594,126 @@ class MulticlassNonMaxSuppressionTest(tf.test.TestCase):
iou_thresh
=
.
5
iou_thresh
=
.
5
max_output_size
=
4
max_output_size
=
4
exp_nms_corners
=
[[[
0
,
10
,
1
,
11
],
exp_nms_corners
=
np
.
array
([[[
0
,
10
,
1
,
11
],
[
0
,
0
,
1
,
1
],
[
0
,
0
,
1
,
1
],
[
0
,
0
,
0
,
0
],
[
0
,
0
,
0
,
0
],
[
0
,
0
,
0
,
0
]],
[
0
,
0
,
0
,
0
]],
[[
0
,
999
,
2
,
1004
],
[[
0
,
999
,
2
,
1004
],
[
0
,
10.1
,
1
,
11.1
],
[
0
,
10.1
,
1
,
11.1
],
[
0
,
100
,
1
,
101
],
[
0
,
100
,
1
,
101
],
[
0
,
0
,
0
,
0
]]]
[
0
,
0
,
0
,
0
]]])
exp_nms_scores
=
[[.
95
,
.
9
,
0
,
0
],
exp_nms_scores
=
np
.
array
([[.
95
,
.
9
,
0
,
0
],
[.
85
,
.
5
,
.
3
,
0
]]
[.
85
,
.
5
,
.
3
,
0
]])
exp_nms_classes
=
[[
0
,
0
,
0
,
0
],
exp_nms_classes
=
np
.
array
([[
0
,
0
,
0
,
0
],
[
1
,
0
,
0
,
0
]]
[
1
,
0
,
0
,
0
]])
exp_nms_masks
=
[[[[
6
,
7
],
[
8
,
9
]],
exp_nms_masks
=
np
.
array
([[[[
6
,
7
],
[
8
,
9
]],
[[
0
,
1
],
[
2
,
3
]],
[[
0
,
1
],
[
2
,
3
]],
[[
0
,
0
],
[
0
,
0
]],
[[
0
,
0
],
[
0
,
0
]],
[[
0
,
0
],
[
0
,
0
]]],
[[
0
,
0
],
[
0
,
0
]]],
[[[
13
,
14
],
[
15
,
16
]],
[[[
13
,
14
],
[
15
,
16
]],
[[
8
,
9
],
[
10
,
11
]],
[[
8
,
9
],
[
10
,
11
]],
[[
10
,
11
],
[
12
,
13
]],
[[
10
,
11
],
[
12
,
13
]],
[[
0
,
0
],
[
0
,
0
]]]]
[[
0
,
0
],
[
0
,
0
]]]])
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
num_detections
)
=
post_processing
.
batch_multiclass_non_max_suppression
(
boxes
,
scores
,
score_thresh
,
iou_thresh
,
max_size_per_class
=
max_output_size
,
max_total_size
=
max_output_size
,
masks
=
masks
)
# Check static shapes
self
.
assertAllEqual
(
nmsed_boxes
.
shape
.
as_list
(),
exp_nms_corners
.
shape
)
self
.
assertAllEqual
(
nmsed_scores
.
shape
.
as_list
(),
exp_nms_scores
.
shape
)
self
.
assertAllEqual
(
nmsed_classes
.
shape
.
as_list
(),
exp_nms_classes
.
shape
)
self
.
assertAllEqual
(
nmsed_masks
.
shape
.
as_list
(),
exp_nms_masks
.
shape
)
self
.
assertEqual
(
num_detections
.
shape
.
as_list
(),
[
2
])
with
self
.
test_session
()
as
sess
:
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
num_detections
)
=
sess
.
run
([
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
num_detections
])
self
.
assertAllClose
(
nmsed_boxes
,
exp_nms_corners
)
self
.
assertAllClose
(
nmsed_scores
,
exp_nms_scores
)
self
.
assertAllClose
(
nmsed_classes
,
exp_nms_classes
)
self
.
assertAllClose
(
num_detections
,
[
2
,
3
])
self
.
assertAllClose
(
nmsed_masks
,
exp_nms_masks
)
def
test_batch_multiclass_nms_with_dynamic_batch_size
(
self
):
boxes_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
(
None
,
None
,
2
,
4
))
scores_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
(
None
,
None
,
2
))
masks_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
(
None
,
None
,
2
,
2
,
2
))
boxes
=
np
.
array
([[[[
0
,
0
,
1
,
1
],
[
0
,
0
,
4
,
5
]],
[[
0
,
0.1
,
1
,
1.1
],
[
0
,
0.1
,
2
,
1.1
]],
[[
0
,
-
0.1
,
1
,
0.9
],
[
0
,
-
0.1
,
1
,
0.9
]],
[[
0
,
10
,
1
,
11
],
[
0
,
10
,
1
,
11
]]],
[[[
0
,
10.1
,
1
,
11.1
],
[
0
,
10.1
,
1
,
11.1
]],
[[
0
,
100
,
1
,
101
],
[
0
,
100
,
1
,
101
]],
[[
0
,
1000
,
1
,
1002
],
[
0
,
999
,
2
,
1004
]],
[[
0
,
1000
,
1
,
1002.1
],
[
0
,
999
,
2
,
1002.7
]]]])
scores
=
np
.
array
([[[.
9
,
0.01
],
[.
75
,
0.05
],
[.
6
,
0.01
],
[.
95
,
0
]],
[[.
5
,
0.01
],
[.
3
,
0.01
],
[.
01
,
.
85
],
[.
01
,
.
5
]]])
masks
=
np
.
array
([[[[[
0
,
1
],
[
2
,
3
]],
[[
1
,
2
],
[
3
,
4
]]],
[[[
2
,
3
],
[
4
,
5
]],
[[
3
,
4
],
[
5
,
6
]]],
[[[
4
,
5
],
[
6
,
7
]],
[[
5
,
6
],
[
7
,
8
]]],
[[[
6
,
7
],
[
8
,
9
]],
[[
7
,
8
],
[
9
,
10
]]]],
[[[[
8
,
9
],
[
10
,
11
]],
[[
9
,
10
],
[
11
,
12
]]],
[[[
10
,
11
],
[
12
,
13
]],
[[
11
,
12
],
[
13
,
14
]]],
[[[
12
,
13
],
[
14
,
15
]],
[[
13
,
14
],
[
15
,
16
]]],
[[[
14
,
15
],
[
16
,
17
]],
[[
15
,
16
],
[
17
,
18
]]]]])
score_thresh
=
0.1
iou_thresh
=
.
5
max_output_size
=
4
exp_nms_corners
=
np
.
array
([[[
0
,
10
,
1
,
11
],
[
0
,
0
,
1
,
1
],
[
0
,
0
,
0
,
0
],
[
0
,
0
,
0
,
0
]],
[[
0
,
999
,
2
,
1004
],
[
0
,
10.1
,
1
,
11.1
],
[
0
,
100
,
1
,
101
],
[
0
,
0
,
0
,
0
]]])
exp_nms_scores
=
np
.
array
([[.
95
,
.
9
,
0
,
0
],
[.
85
,
.
5
,
.
3
,
0
]])
exp_nms_classes
=
np
.
array
([[
0
,
0
,
0
,
0
],
[
1
,
0
,
0
,
0
]])
exp_nms_masks
=
np
.
array
([[[[
6
,
7
],
[
8
,
9
]],
[[
0
,
1
],
[
2
,
3
]],
[[
0
,
0
],
[
0
,
0
]],
[[
0
,
0
],
[
0
,
0
]]],
[[[
13
,
14
],
[
15
,
16
]],
[[
8
,
9
],
[
10
,
11
]],
[[
10
,
11
],
[
12
,
13
]],
[[
0
,
0
],
[
0
,
0
]]]])
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
num_detections
)
=
post_processing
.
batch_multiclass_non_max_suppression
(
boxes_placeholder
,
scores_placeholder
,
score_thresh
,
iou_thresh
,
max_size_per_class
=
max_output_size
,
max_total_size
=
max_output_size
,
masks
=
masks_placeholder
)
# Check static shapes
self
.
assertAllEqual
(
nmsed_boxes
.
shape
.
as_list
(),
[
None
,
4
,
4
])
self
.
assertAllEqual
(
nmsed_scores
.
shape
.
as_list
(),
[
None
,
4
])
self
.
assertAllEqual
(
nmsed_classes
.
shape
.
as_list
(),
[
None
,
4
])
self
.
assertAllEqual
(
nmsed_masks
.
shape
.
as_list
(),
[
None
,
4
,
2
,
2
])
self
.
assertEqual
(
num_detections
.
shape
.
as_list
(),
[
None
])
nms_dict
=
post_processing
.
batch_multiclass_non_max_suppression
(
boxes
,
scores
,
score_thresh
,
iou_thresh
,
max_size_per_class
=
max_output_size
,
max_total_size
=
max_output_size
,
masks
=
masks
)
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
nms_output
=
sess
.
run
(
nms_dict
)
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
self
.
assertAllClose
(
nms_output
[
'detection_boxes'
],
exp_nms_corners
)
num_detections
)
=
sess
.
run
([
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
self
.
assertAllClose
(
nms_output
[
'detection_scores'
],
exp_nms_scores
)
nmsed_masks
,
num_detections
],
self
.
assertAllClose
(
nms_output
[
'detection_classes'
],
exp_nms_classes
)
feed_dict
=
{
boxes_placeholder
:
boxes
,
self
.
assertAllClose
(
nms_output
[
'num_detections'
],
[
2
,
3
])
scores_placeholder
:
scores
,
self
.
assertAllClose
(
nms_output
[
'detection_masks'
],
exp_nms_masks
)
masks_placeholder
:
masks
})
self
.
assertAllClose
(
nmsed_boxes
,
exp_nms_corners
)
self
.
assertAllClose
(
nmsed_scores
,
exp_nms_scores
)
self
.
assertAllClose
(
nmsed_classes
,
exp_nms_classes
)
self
.
assertAllClose
(
num_detections
,
[
2
,
3
])
self
.
assertAllClose
(
nmsed_masks
,
exp_nms_masks
)
def
test_batch_multiclass_nms_with_masks_and_num_valid_boxes
(
self
):
def
test_batch_multiclass_nms_with_masks_and_num_valid_boxes
(
self
):
boxes
=
tf
.
constant
([[[[
0
,
0
,
1
,
1
],
[
0
,
0
,
4
,
5
]],
boxes
=
tf
.
constant
([[[[
0
,
0
,
1
,
1
],
[
0
,
0
,
4
,
5
]],
...
@@ -656,17 +764,21 @@ class MulticlassNonMaxSuppressionTest(tf.test.TestCase):
...
@@ -656,17 +764,21 @@ class MulticlassNonMaxSuppressionTest(tf.test.TestCase):
[[
0
,
0
],
[
0
,
0
]],
[[
0
,
0
],
[
0
,
0
]],
[[
0
,
0
],
[
0
,
0
]]]]
[[
0
,
0
],
[
0
,
0
]]]]
nms_dict
=
post_processing
.
batch_multiclass_non_max_suppression
(
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
boxes
,
scores
,
score_thresh
,
iou_thresh
,
num_detections
)
=
post_processing
.
batch_multiclass_non_max_suppression
(
max_size_per_class
=
max_output_size
,
max_total_size
=
max_output_size
,
boxes
,
scores
,
score_thresh
,
iou_thresh
,
num_valid_boxes
=
num_valid_boxes
,
masks
=
masks
)
max_size_per_class
=
max_output_size
,
max_total_size
=
max_output_size
,
num_valid_boxes
=
num_valid_boxes
,
masks
=
masks
)
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
nms_output
=
sess
.
run
(
nms_dict
)
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
self
.
assertAllClose
(
nms_output
[
'detection_boxes'
],
exp_nms_corners
)
num_detections
)
=
sess
.
run
([
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
self
.
assertAllClose
(
nms_output
[
'detection_scores'
],
exp_nms_scores
)
nmsed_masks
,
num_detections
])
self
.
assertAllClose
(
nms_output
[
'detection_classes'
],
exp_nms_classes
)
self
.
assertAllClose
(
nmsed_boxes
,
exp_nms_corners
)
self
.
assertAllClose
(
nms_output
[
'num_detections'
],
[
1
,
1
])
self
.
assertAllClose
(
nmsed_scores
,
exp_nms_scores
)
self
.
assertAllClose
(
nms_output
[
'detection_masks'
],
exp_nms_masks
)
self
.
assertAllClose
(
nmsed_classes
,
exp_nms_classes
)
self
.
assertAllClose
(
num_detections
,
[
1
,
1
])
self
.
assertAllClose
(
nmsed_masks
,
exp_nms_masks
)
if
__name__
==
'__main__'
:
if
__name__
==
'__main__'
:
...
...
object_detection/core/preprocessor.py
View file @
582bf927
...
@@ -1255,6 +1255,82 @@ def random_resize_method(image, target_size):
...
@@ -1255,6 +1255,82 @@ def random_resize_method(image, target_size):
return
resized_image
return
resized_image
def
_compute_new_static_size
(
image
,
min_dimension
,
max_dimension
):
"""Compute new static shape for resize_to_range method."""
image_shape
=
image
.
get_shape
().
as_list
()
orig_height
=
image_shape
[
0
]
orig_width
=
image_shape
[
1
]
orig_min_dim
=
min
(
orig_height
,
orig_width
)
# Calculates the larger of the possible sizes
large_scale_factor
=
min_dimension
/
float
(
orig_min_dim
)
# Scaling orig_(height|width) by large_scale_factor will make the smaller
# dimension equal to min_dimension, save for floating point rounding errors.
# For reasonably-sized images, taking the nearest integer will reliably
# eliminate this error.
large_height
=
int
(
round
(
orig_height
*
large_scale_factor
))
large_width
=
int
(
round
(
orig_width
*
large_scale_factor
))
large_size
=
[
large_height
,
large_width
]
if
max_dimension
:
# Calculates the smaller of the possible sizes, use that if the larger
# is too big.
orig_max_dim
=
max
(
orig_height
,
orig_width
)
small_scale_factor
=
max_dimension
/
float
(
orig_max_dim
)
# Scaling orig_(height|width) by small_scale_factor will make the larger
# dimension equal to max_dimension, save for floating point rounding
# errors. For reasonably-sized images, taking the nearest integer will
# reliably eliminate this error.
small_height
=
int
(
round
(
orig_height
*
small_scale_factor
))
small_width
=
int
(
round
(
orig_width
*
small_scale_factor
))
small_size
=
[
small_height
,
small_width
]
new_size
=
large_size
if
max
(
large_size
)
>
max_dimension
:
new_size
=
small_size
else
:
new_size
=
large_size
return
tf
.
constant
(
new_size
)
def
_compute_new_dynamic_size
(
image
,
min_dimension
,
max_dimension
):
"""Compute new dynamic shape for resize_to_range method."""
image_shape
=
tf
.
shape
(
image
)
orig_height
=
tf
.
to_float
(
image_shape
[
0
])
orig_width
=
tf
.
to_float
(
image_shape
[
1
])
orig_min_dim
=
tf
.
minimum
(
orig_height
,
orig_width
)
# Calculates the larger of the possible sizes
min_dimension
=
tf
.
constant
(
min_dimension
,
dtype
=
tf
.
float32
)
large_scale_factor
=
min_dimension
/
orig_min_dim
# Scaling orig_(height|width) by large_scale_factor will make the smaller
# dimension equal to min_dimension, save for floating point rounding errors.
# For reasonably-sized images, taking the nearest integer will reliably
# eliminate this error.
large_height
=
tf
.
to_int32
(
tf
.
round
(
orig_height
*
large_scale_factor
))
large_width
=
tf
.
to_int32
(
tf
.
round
(
orig_width
*
large_scale_factor
))
large_size
=
tf
.
stack
([
large_height
,
large_width
])
if
max_dimension
:
# Calculates the smaller of the possible sizes, use that if the larger
# is too big.
orig_max_dim
=
tf
.
maximum
(
orig_height
,
orig_width
)
max_dimension
=
tf
.
constant
(
max_dimension
,
dtype
=
tf
.
float32
)
small_scale_factor
=
max_dimension
/
orig_max_dim
# Scaling orig_(height|width) by small_scale_factor will make the larger
# dimension equal to max_dimension, save for floating point rounding
# errors. For reasonably-sized images, taking the nearest integer will
# reliably eliminate this error.
small_height
=
tf
.
to_int32
(
tf
.
round
(
orig_height
*
small_scale_factor
))
small_width
=
tf
.
to_int32
(
tf
.
round
(
orig_width
*
small_scale_factor
))
small_size
=
tf
.
stack
([
small_height
,
small_width
])
new_size
=
tf
.
cond
(
tf
.
to_float
(
tf
.
reduce_max
(
large_size
))
>
max_dimension
,
lambda
:
small_size
,
lambda
:
large_size
)
else
:
new_size
=
large_size
return
new_size
def
resize_to_range
(
image
,
def
resize_to_range
(
image
,
masks
=
None
,
masks
=
None
,
min_dimension
=
None
,
min_dimension
=
None
,
...
@@ -1295,64 +1371,22 @@ def resize_to_range(image,
...
@@ -1295,64 +1371,22 @@ def resize_to_range(image,
raise
ValueError
(
'Image should be 3D tensor'
)
raise
ValueError
(
'Image should be 3D tensor'
)
with
tf
.
name_scope
(
'ResizeToRange'
,
values
=
[
image
,
min_dimension
]):
with
tf
.
name_scope
(
'ResizeToRange'
,
values
=
[
image
,
min_dimension
]):
image_shape
=
tf
.
shape
(
image
)
if
image
.
get_shape
().
is_fully_defined
():
orig_height
=
tf
.
to_float
(
image_shape
[
0
])
new_size
=
_compute_new_static_size
(
image
,
min_dimension
,
orig_width
=
tf
.
to_float
(
image_shape
[
1
])
max_dimension
)
orig_min_dim
=
tf
.
minimum
(
orig_height
,
orig_width
)
# Calculates the larger of the possible sizes
min_dimension
=
tf
.
constant
(
min_dimension
,
dtype
=
tf
.
float32
)
large_scale_factor
=
min_dimension
/
orig_min_dim
# Scaling orig_(height|width) by large_scale_factor will make the smaller
# dimension equal to min_dimension, save for floating point rounding errors.
# For reasonably-sized images, taking the nearest integer will reliably
# eliminate this error.
large_height
=
tf
.
to_int32
(
tf
.
round
(
orig_height
*
large_scale_factor
))
large_width
=
tf
.
to_int32
(
tf
.
round
(
orig_width
*
large_scale_factor
))
large_size
=
tf
.
stack
([
large_height
,
large_width
])
if
max_dimension
:
# Calculates the smaller of the possible sizes, use that if the larger
# is too big.
orig_max_dim
=
tf
.
maximum
(
orig_height
,
orig_width
)
max_dimension
=
tf
.
constant
(
max_dimension
,
dtype
=
tf
.
float32
)
small_scale_factor
=
max_dimension
/
orig_max_dim
# Scaling orig_(height|width) by small_scale_factor will make the larger
# dimension equal to max_dimension, save for floating point rounding
# errors. For reasonably-sized images, taking the nearest integer will
# reliably eliminate this error.
small_height
=
tf
.
to_int32
(
tf
.
round
(
orig_height
*
small_scale_factor
))
small_width
=
tf
.
to_int32
(
tf
.
round
(
orig_width
*
small_scale_factor
))
small_size
=
tf
.
stack
([
small_height
,
small_width
])
new_size
=
tf
.
cond
(
tf
.
to_float
(
tf
.
reduce_max
(
large_size
))
>
max_dimension
,
lambda
:
small_size
,
lambda
:
large_size
)
else
:
else
:
new_size
=
large_size
new_size
=
_compute_new_dynamic_size
(
image
,
min_dimension
,
max_dimension
)
new_image
=
tf
.
image
.
resize_images
(
image
,
new_size
,
new_image
=
tf
.
image
.
resize_images
(
image
,
new_size
,
align_corners
=
align_corners
)
align_corners
=
align_corners
)
result
=
new_image
result
=
new_image
if
masks
is
not
None
:
if
masks
is
not
None
:
num_instances
=
tf
.
shape
(
masks
)[
0
]
new_masks
=
tf
.
expand_dims
(
masks
,
3
)
new_masks
=
tf
.
image
.
resize_nearest_neighbor
(
new_masks
,
new_size
,
def
resize_masks_branch
():
align_corners
=
align_corners
)
new_masks
=
tf
.
expand_dims
(
masks
,
3
)
new_masks
=
tf
.
squeeze
(
new_masks
,
3
)
new_masks
=
tf
.
image
.
resize_nearest_neighbor
(
result
=
[
new_image
,
new_masks
]
new_masks
,
new_size
,
align_corners
=
align_corners
)
new_masks
=
tf
.
squeeze
(
new_masks
,
axis
=
3
)
return
new_masks
def
reshape_masks_branch
():
new_masks
=
tf
.
reshape
(
masks
,
[
0
,
new_size
[
0
],
new_size
[
1
]])
return
new_masks
masks
=
tf
.
cond
(
num_instances
>
0
,
resize_masks_branch
,
reshape_masks_branch
)
result
=
[
new_image
,
masks
]
return
result
return
result
...
...
object_detection/core/preprocessor_test.py
View file @
582bf927
...
@@ -1395,7 +1395,7 @@ class PreprocessorTest(tf.test.TestCase):
...
@@ -1395,7 +1395,7 @@ class PreprocessorTest(tf.test.TestCase):
self
.
assertAllEqual
(
expected_images_shape_
,
self
.
assertAllEqual
(
expected_images_shape_
,
resized_images_shape_
)
resized_images_shape_
)
def
testResizeToRange
(
self
):
def
testResizeToRange
PreservesStaticSpatialShape
(
self
):
"""Tests image resizing, checking output sizes."""
"""Tests image resizing, checking output sizes."""
in_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
],
[
15
,
50
,
3
]]
in_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
],
[
15
,
50
,
3
]]
min_dim
=
50
min_dim
=
50
...
@@ -1406,13 +1406,27 @@ class PreprocessorTest(tf.test.TestCase):
...
@@ -1406,13 +1406,27 @@ class PreprocessorTest(tf.test.TestCase):
in_image
=
tf
.
random_uniform
(
in_shape
)
in_image
=
tf
.
random_uniform
(
in_shape
)
out_image
=
preprocessor
.
resize_to_range
(
out_image
=
preprocessor
.
resize_to_range
(
in_image
,
min_dimension
=
min_dim
,
max_dimension
=
max_dim
)
in_image
,
min_dimension
=
min_dim
,
max_dimension
=
max_dim
)
out_image_shape
=
tf
.
shape
(
out_image
)
self
.
assertAllEqual
(
out_image
.
get_shape
().
as_list
(),
expected_shape
)
def
testResizeToRangeWithDynamicSpatialShape
(
self
):
"""Tests image resizing, checking output sizes."""
in_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
],
[
15
,
50
,
3
]]
min_dim
=
50
max_dim
=
100
expected_shape_list
=
[[
75
,
50
,
3
],
[
50
,
100
,
3
],
[
30
,
100
,
3
]]
for
in_shape
,
expected_shape
in
zip
(
in_shape_list
,
expected_shape_list
):
in_image
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
(
None
,
None
,
3
))
out_image
=
preprocessor
.
resize_to_range
(
in_image
,
min_dimension
=
min_dim
,
max_dimension
=
max_dim
)
out_image_shape
=
tf
.
shape
(
out_image
)
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
out_image_shape
=
sess
.
run
(
out_image_shape
)
out_image_shape
=
sess
.
run
(
out_image_shape
,
feed_dict
=
{
in_image
:
np
.
random
.
randn
(
*
in_shape
)})
self
.
assertAllEqual
(
out_image_shape
,
expected_shape
)
self
.
assertAllEqual
(
out_image_shape
,
expected_shape
)
def
testResizeToRangeWithMasks
(
self
):
def
testResizeToRangeWithMasks
PreservesStaticSpatialShape
(
self
):
"""Tests image resizing, checking output sizes."""
"""Tests image resizing, checking output sizes."""
in_image_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
]]
in_image_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
]]
in_masks_shape_list
=
[[
15
,
60
,
40
],
[
10
,
15
,
30
]]
in_masks_shape_list
=
[[
15
,
60
,
40
],
[
10
,
15
,
30
]]
...
@@ -1430,30 +1444,25 @@ class PreprocessorTest(tf.test.TestCase):
...
@@ -1430,30 +1444,25 @@ class PreprocessorTest(tf.test.TestCase):
in_masks
=
tf
.
random_uniform
(
in_masks_shape
)
in_masks
=
tf
.
random_uniform
(
in_masks_shape
)
out_image
,
out_masks
=
preprocessor
.
resize_to_range
(
out_image
,
out_masks
=
preprocessor
.
resize_to_range
(
in_image
,
in_masks
,
min_dimension
=
min_dim
,
max_dimension
=
max_dim
)
in_image
,
in_masks
,
min_dimension
=
min_dim
,
max_dimension
=
max_dim
)
out_image_shape
=
tf
.
shape
(
out_image
)
self
.
assertAllEqual
(
out_masks
.
get_shape
().
as_list
(),
expected_mask_shape
)
out_masks_shape
=
tf
.
shape
(
out_masks
)
self
.
assertAllEqual
(
out_image
.
get_shape
().
as_list
(),
expected_image_shape
)
with
self
.
test_session
()
as
sess
:
out_image_shape
,
out_masks_shape
=
sess
.
run
(
[
out_image_shape
,
out_masks_shape
])
self
.
assertAllEqual
(
out_image_shape
,
expected_image_shape
)
self
.
assertAllEqual
(
out_masks_shape
,
expected_mask_shape
)
def
testResizeToRangeWith
NoInstanceMask
(
self
):
def
testResizeToRangeWith
MasksAndDynamicSpatialShape
(
self
):
"""Tests image resizing, checking output sizes."""
"""Tests image resizing, checking output sizes."""
in_image_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
]]
in_image_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
]]
in_masks_shape_list
=
[[
0
,
60
,
40
],
[
0
,
15
,
30
]]
in_masks_shape_list
=
[[
15
,
60
,
40
],
[
1
0
,
15
,
30
]]
min_dim
=
50
min_dim
=
50
max_dim
=
100
max_dim
=
100
expected_image_shape_list
=
[[
75
,
50
,
3
],
[
50
,
100
,
3
]]
expected_image_shape_list
=
[[
75
,
50
,
3
],
[
50
,
100
,
3
]]
expected_masks_shape_list
=
[[
0
,
75
,
50
],
[
0
,
50
,
100
]]
expected_masks_shape_list
=
[[
15
,
75
,
50
],
[
1
0
,
50
,
100
]]
for
(
in_image_shape
,
expected_image_shape
,
in_masks_shape
,
for
(
in_image_shape
,
expected_image_shape
,
in_masks_shape
,
expected_mask_shape
)
in
zip
(
in_image_shape_list
,
expected_mask_shape
)
in
zip
(
in_image_shape_list
,
expected_image_shape_list
,
expected_image_shape_list
,
in_masks_shape_list
,
in_masks_shape_list
,
expected_masks_shape_list
):
expected_masks_shape_list
):
in_image
=
tf
.
random_uniform
(
in_image_shape
)
in_image
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
(
None
,
None
,
3
))
in_masks
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
(
None
,
None
,
None
))
in_masks
=
tf
.
random_uniform
(
in_masks_shape
)
in_masks
=
tf
.
random_uniform
(
in_masks_shape
)
out_image
,
out_masks
=
preprocessor
.
resize_to_range
(
out_image
,
out_masks
=
preprocessor
.
resize_to_range
(
in_image
,
in_masks
,
min_dimension
=
min_dim
,
max_dimension
=
max_dim
)
in_image
,
in_masks
,
min_dimension
=
min_dim
,
max_dimension
=
max_dim
)
...
@@ -1462,38 +1471,15 @@ class PreprocessorTest(tf.test.TestCase):
...
@@ -1462,38 +1471,15 @@ class PreprocessorTest(tf.test.TestCase):
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
out_image_shape
,
out_masks_shape
=
sess
.
run
(
out_image_shape
,
out_masks_shape
=
sess
.
run
(
[
out_image_shape
,
out_masks_shape
])
[
out_image_shape
,
out_masks_shape
],
self
.
assertAllEqual
(
out_image_shape
,
expected_image_shape
)
feed_dict
=
{
self
.
assertAllEqual
(
out_masks_shape
,
expected_mask_shape
)
in_image
:
np
.
random
.
randn
(
*
in_image_shape
),
in_masks
:
np
.
random
.
randn
(
*
in_masks_shape
)
def
testResizeImageWithMasks
(
self
):
})
"""Tests image resizing, checking output sizes."""
in_image_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
]]
in_masks_shape_list
=
[[
15
,
60
,
40
],
[
10
,
15
,
30
]]
height
=
50
width
=
100
expected_image_shape_list
=
[[
50
,
100
,
3
],
[
50
,
100
,
3
]]
expected_masks_shape_list
=
[[
15
,
50
,
100
],
[
10
,
50
,
100
]]
for
(
in_image_shape
,
expected_image_shape
,
in_masks_shape
,
expected_mask_shape
)
in
zip
(
in_image_shape_list
,
expected_image_shape_list
,
in_masks_shape_list
,
expected_masks_shape_list
):
in_image
=
tf
.
random_uniform
(
in_image_shape
)
in_masks
=
tf
.
random_uniform
(
in_masks_shape
)
out_image
,
out_masks
=
preprocessor
.
resize_image
(
in_image
,
in_masks
,
new_height
=
height
,
new_width
=
width
)
out_image_shape
=
tf
.
shape
(
out_image
)
out_masks_shape
=
tf
.
shape
(
out_masks
)
with
self
.
test_session
()
as
sess
:
out_image_shape
,
out_masks_shape
=
sess
.
run
(
[
out_image_shape
,
out_masks_shape
])
self
.
assertAllEqual
(
out_image_shape
,
expected_image_shape
)
self
.
assertAllEqual
(
out_image_shape
,
expected_image_shape
)
self
.
assertAllEqual
(
out_masks_shape
,
expected_mask_shape
)
self
.
assertAllEqual
(
out_masks_shape
,
expected_mask_shape
)
def
testResize
Ima
geWith
No
InstanceMask
(
self
):
def
testResize
ToRan
geWithInstanceMask
sTensorOfSizeZero
(
self
):
"""Tests image resizing, checking output sizes."""
"""Tests image resizing, checking output sizes."""
in_image_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
]]
in_image_shape_list
=
[[
60
,
40
,
3
],
[
15
,
30
,
3
]]
in_masks_shape_list
=
[[
0
,
60
,
40
],
[
0
,
15
,
30
]]
in_masks_shape_list
=
[[
0
,
60
,
40
],
[
0
,
15
,
30
]]
...
...
object_detection/export_inference_graph.py
View file @
582bf927
...
@@ -16,16 +16,19 @@
...
@@ -16,16 +16,19 @@
r
"""Tool to export an object detection model for inference.
r
"""Tool to export an object detection model for inference.
Prepares an object detection tensorflow graph for inference using model
Prepares an object detection tensorflow graph for inference using model
configuration and an optional trained checkpoint. Outputs either an inference
configuration and an optional trained checkpoint. Outputs inference
graph or a SavedModel (https://tensorflow.github.io/serving/serving_basic.html).
graph, associated checkpoint files, a frozen inference graph and a
SavedModel (https://tensorflow.github.io/serving/serving_basic.html).
The inference graph contains one of three input nodes depending on the user
The inference graph contains one of three input nodes depending on the user
specified option.
specified option.
* `image_tensor`: Accepts a uint8 4-D tensor of shape [1, None, None, 3]
* `image_tensor`: Accepts a uint8 4-D tensor of shape [None, None, None, 3]
* `encoded_image_string_tensor`: Accepts a scalar string tensor of encoded PNG
* `encoded_image_string_tensor`: Accepts a 1-D string tensor of shape [None]
or JPEG image.
containing encoded PNG or JPEG images. Image resolutions are expected to be
* `tf_example`: Accepts a serialized TFExample proto. The batch size in this
the same if more than 1 image is provided.
case is always 1.
* `tf_example`: Accepts a 1-D string tensor of shape [None] containing
serialized TFExample protos. Image resolutions are expected to be the same
if more than 1 image is provided.
and the following output nodes returned by the model.postprocess(..):
and the following output nodes returned by the model.postprocess(..):
* `num_detections`: Outputs float32 tensors of the form [batch]
* `num_detections`: Outputs float32 tensors of the form [batch]
...
@@ -41,23 +44,27 @@ and the following output nodes returned by the model.postprocess(..):
...
@@ -41,23 +44,27 @@ and the following output nodes returned by the model.postprocess(..):
masks for each box if its present in the dictionary of postprocessed
masks for each box if its present in the dictionary of postprocessed
tensors returned by the model.
tensors returned by the model.
Note that currently `batch` is always 1, but we will support `batch` > 1 in
Notes:
the future.
* This tool uses `use_moving_averages` from eval_config to decide which
weights to freeze.
Optionally, one can freeze the graph by converting the weights in the provided
checkpoint as graph constants thereby eliminating the need to use a checkpoint
file during inference.
Note that this tool uses `use_moving_averages` from eval_config to decide
which weights to freeze.
Example Usage:
Example Usage:
--------------
--------------
python export_inference_graph \
python export_inference_graph \
--input_type image_tensor \
--input_type image_tensor \
--pipeline_config_path path/to/ssd_inception_v2.config \
--pipeline_config_path path/to/ssd_inception_v2.config \
--checkpoint_path path/to/model-ckpt \
--trained_checkpoint_prefix path/to/model.ckpt \
--inference_graph_path path/to/inference_graph.pb
--output_directory path/to/exported_model_directory
The expected output would be in the directory
path/to/exported_model_directory (which is created if it does not exist)
with contents:
- graph.pbtxt
- model.ckpt.data-00000-of-00001
- model.ckpt.info
- model.ckpt.meta
- frozen_inference_graph.pb
+ saved_model (a directory)
"""
"""
import
tensorflow
as
tf
import
tensorflow
as
tf
from
google.protobuf
import
text_format
from
google.protobuf
import
text_format
...
@@ -70,31 +77,29 @@ flags = tf.app.flags
...
@@ -70,31 +77,29 @@ flags = tf.app.flags
flags
.
DEFINE_string
(
'input_type'
,
'image_tensor'
,
'Type of input node. Can be '
flags
.
DEFINE_string
(
'input_type'
,
'image_tensor'
,
'Type of input node. Can be '
'one of [`image_tensor`, `encoded_image_string_tensor`, '
'one of [`image_tensor`, `encoded_image_string_tensor`, '
'`tf_example`]'
)
'`tf_example`]'
)
flags
.
DEFINE_string
(
'pipeline_config_path'
,
''
,
flags
.
DEFINE_string
(
'pipeline_config_path'
,
None
,
'Path to a pipeline_pb2.TrainEvalPipelineConfig config '
'Path to a pipeline_pb2.TrainEvalPipelineConfig config '
'file.'
)
'file.'
)
flags
.
DEFINE_string
(
'checkpoint_path'
,
''
,
'Optional path to checkpoint file. '
flags
.
DEFINE_string
(
'trained_checkpoint_prefix'
,
None
,
'If provided, bakes the weights from the checkpoint into '
'Path to trained checkpoint, typically of the form '
'the graph.'
)
'path/to/model.ckpt'
)
flags
.
DEFINE_string
(
'inference_graph_path'
,
''
,
'Path to write the output '
flags
.
DEFINE_string
(
'output_directory'
,
None
,
'Path to write outputs.'
)
'inference graph.'
)
flags
.
DEFINE_bool
(
'export_as_saved_model'
,
False
,
'Whether the exported graph '
'should be saved as a SavedModel'
)
FLAGS
=
flags
.
FLAGS
FLAGS
=
flags
.
FLAGS
def
main
(
_
):
def
main
(
_
):
assert
FLAGS
.
pipeline_config_path
,
'TrainEvalPipelineConfig missing.'
assert
FLAGS
.
pipeline_config_path
,
'`pipeline_config_path` is missing'
assert
FLAGS
.
inference_graph_path
,
'Inference graph path missing.'
assert
FLAGS
.
trained_checkpoint_prefix
,
(
assert
FLAGS
.
input_type
,
'Input type missing.'
'`trained_checkpoint_prefix` is missing'
)
assert
FLAGS
.
output_directory
,
'`output_directory` is missing'
pipeline_config
=
pipeline_pb2
.
TrainEvalPipelineConfig
()
pipeline_config
=
pipeline_pb2
.
TrainEvalPipelineConfig
()
with
tf
.
gfile
.
GFile
(
FLAGS
.
pipeline_config_path
,
'r'
)
as
f
:
with
tf
.
gfile
.
GFile
(
FLAGS
.
pipeline_config_path
,
'r'
)
as
f
:
text_format
.
Merge
(
f
.
read
(),
pipeline_config
)
text_format
.
Merge
(
f
.
read
(),
pipeline_config
)
exporter
.
export_inference_graph
(
FLAGS
.
input_type
,
pipeline_config
,
exporter
.
export_inference_graph
(
FLAGS
.
checkpoint_path
,
FLAGS
.
input_type
,
pipeline_config
,
FLAGS
.
trained_checkpoint_prefix
,
FLAGS
.
inference_graph_path
,
FLAGS
.
output_directory
)
FLAGS
.
export_as_saved_model
)
if
__name__
==
'__main__'
:
if
__name__
==
'__main__'
:
...
...
object_detection/exporter.py
View file @
582bf927
...
@@ -17,6 +17,7 @@
...
@@ -17,6 +17,7 @@
import
logging
import
logging
import
os
import
os
import
tensorflow
as
tf
import
tensorflow
as
tf
from
tensorflow.core.protobuf
import
rewriter_config_pb2
from
tensorflow.python
import
pywrap_tensorflow
from
tensorflow.python
import
pywrap_tensorflow
from
tensorflow.python.client
import
session
from
tensorflow.python.client
import
session
from
tensorflow.python.framework
import
graph_util
from
tensorflow.python.framework
import
graph_util
...
@@ -42,6 +43,7 @@ def freeze_graph_with_def_protos(
...
@@ -42,6 +43,7 @@ def freeze_graph_with_def_protos(
filename_tensor_name
,
filename_tensor_name
,
clear_devices
,
clear_devices
,
initializer_nodes
,
initializer_nodes
,
optimize_graph
=
False
,
variable_names_blacklist
=
''
):
variable_names_blacklist
=
''
):
"""Converts all variables in a graph and checkpoint into constants."""
"""Converts all variables in a graph and checkpoint into constants."""
del
restore_op_name
,
filename_tensor_name
# Unused by updated loading code.
del
restore_op_name
,
filename_tensor_name
# Unused by updated loading code.
...
@@ -61,86 +63,106 @@ def freeze_graph_with_def_protos(
...
@@ -61,86 +63,106 @@ def freeze_graph_with_def_protos(
for
node
in
input_graph_def
.
node
:
for
node
in
input_graph_def
.
node
:
node
.
device
=
''
node
.
device
=
''
_
=
importer
.
import_graph_def
(
input_graph_def
,
name
=
''
)
with
tf
.
Graph
().
as_default
():
tf
.
import_graph_def
(
input_graph_def
,
name
=
''
)
with
session
.
Session
()
as
sess
:
if
input_saver_def
:
if
optimize_graph
:
saver
=
saver_lib
.
Saver
(
saver_def
=
input_saver_def
)
logging
.
info
(
'Graph Rewriter optimizations enabled'
)
saver
.
restore
(
sess
,
input_checkpoint
)
rewrite_options
=
rewriter_config_pb2
.
RewriterConfig
(
optimize_tensor_layout
=
True
)
rewrite_options
.
optimizers
.
append
(
'pruning'
)
rewrite_options
.
optimizers
.
append
(
'constfold'
)
rewrite_options
.
optimizers
.
append
(
'layout'
)
graph_options
=
tf
.
GraphOptions
(
rewrite_options
=
rewrite_options
,
infer_shapes
=
True
)
else
:
else
:
var_list
=
{}
logging
.
info
(
'Graph Rewriter optimizations disabled'
)
reader
=
pywrap_tensorflow
.
NewCheckpointReader
(
input_checkpoint
)
graph_options
=
tf
.
GraphOptions
()
var_to_shape_map
=
reader
.
get_variable_to_shape_map
()
config
=
tf
.
ConfigProto
(
graph_options
=
graph_options
)
for
key
in
var_to_shape_map
:
with
session
.
Session
(
config
=
config
)
as
sess
:
try
:
if
input_saver_def
:
tensor
=
sess
.
graph
.
get_tensor_by_name
(
key
+
':0'
)
saver
=
saver_lib
.
Saver
(
saver_def
=
input_saver_def
)
except
KeyError
:
saver
.
restore
(
sess
,
input_checkpoint
)
# This tensor doesn't exist in the graph (for example it's
else
:
# 'global_step' or a similar housekeeping element) so skip it.
var_list
=
{}
continue
reader
=
pywrap_tensorflow
.
NewCheckpointReader
(
input_checkpoint
)
var_list
[
key
]
=
tensor
var_to_shape_map
=
reader
.
get_variable_to_shape_map
()
saver
=
saver_lib
.
Saver
(
var_list
=
var_list
)
for
key
in
var_to_shape_map
:
saver
.
restore
(
sess
,
input_checkpoint
)
try
:
if
initializer_nodes
:
tensor
=
sess
.
graph
.
get_tensor_by_name
(
key
+
':0'
)
sess
.
run
(
initializer_nodes
)
except
KeyError
:
# This tensor doesn't exist in the graph (for example it's
variable_names_blacklist
=
(
variable_names_blacklist
.
split
(
','
)
if
# 'global_step' or a similar housekeeping element) so skip it.
variable_names_blacklist
else
None
)
continue
output_graph_def
=
graph_util
.
convert_variables_to_constants
(
var_list
[
key
]
=
tensor
sess
,
saver
=
saver_lib
.
Saver
(
var_list
=
var_list
)
input_graph_def
,
saver
.
restore
(
sess
,
input_checkpoint
)
output_node_names
.
split
(
','
),
if
initializer_nodes
:
variable_names_blacklist
=
variable_names_blacklist
)
sess
.
run
(
initializer_nodes
)
variable_names_blacklist
=
(
variable_names_blacklist
.
split
(
','
)
if
variable_names_blacklist
else
None
)
output_graph_def
=
graph_util
.
convert_variables_to_constants
(
sess
,
input_graph_def
,
output_node_names
.
split
(
','
),
variable_names_blacklist
=
variable_names_blacklist
)
return
output_graph_def
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
(
def
_image_tensor_input_placeholder
():
input_graph_def
=
inference_graph_def
,
"""Returns placeholder and input node that accepts a batch of uint8 images."""
input_saver_def
=
saver
.
as_saver_def
(),
input_tensor
=
tf
.
placeholder
(
dtype
=
tf
.
uint8
,
input_checkpoint
=
input_checkpoint
,
shape
=
(
None
,
None
,
None
,
3
),
output_node_names
=
output_node_names
,
name
=
'image_tensor'
)
restore_op_name
=
'save/restore_all'
,
return
input_tensor
,
input_tensor
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
():
def
_tf_example_input_placeholder
():
tf_example_placeholder
=
tf
.
placeholder
(
"""Returns input that accepts a batch of strings with tf examples.
tf
.
string
,
shape
=
[],
name
=
'tf_example'
)
tensor_dict
=
tf_example_decoder
.
TfExampleDecoder
().
decode
(
tf_example_placeholder
)
image
=
tensor_dict
[
fields
.
InputDataFields
.
image
]
return
tf
.
expand_dims
(
image
,
axis
=
0
)
Returns:
def
_image_tensor_input_placeholder
():
a tuple of placeholder and input nodes that output decoded images.
return
tf
.
placeholder
(
dtype
=
tf
.
uint8
,
"""
shape
=
(
1
,
None
,
None
,
3
),
batch_tf_example_placeholder
=
tf
.
placeholder
(
name
=
'image_tensor'
)
tf
.
string
,
shape
=
[
None
],
name
=
'tf_example'
)
def
decode
(
tf_example_string_tensor
):
tensor_dict
=
tf_example_decoder
.
TfExampleDecoder
().
decode
(
tf_example_string_tensor
)
image_tensor
=
tensor_dict
[
fields
.
InputDataFields
.
image
]
return
image_tensor
return
(
batch_tf_example_placeholder
,
tf
.
map_fn
(
decode
,
elems
=
batch_tf_example_placeholder
,
dtype
=
tf
.
uint8
,
parallel_iterations
=
32
,
back_prop
=
False
))
def
_encoded_image_string_tensor_input_placeholder
():
def
_encoded_image_string_tensor_input_placeholder
():
image_str
=
tf
.
placeholder
(
dtype
=
tf
.
string
,
"""Returns input that accepts a batch of PNG or JPEG strings.
shape
=
[],
name
=
'encoded_image_string_tensor'
)
Returns:
image_tensor
=
tf
.
image
.
decode_image
(
image_str
,
channels
=
3
)
a tuple of placeholder and input nodes that output decoded images.
image_tensor
.
set_shape
((
None
,
None
,
3
))
"""
return
tf
.
expand_dims
(
image_tensor
,
axis
=
0
)
batch_image_str_placeholder
=
tf
.
placeholder
(
dtype
=
tf
.
string
,
shape
=
[
None
],
name
=
'encoded_image_string_tensor'
)
def
decode
(
encoded_image_string_tensor
):
image_tensor
=
tf
.
image
.
decode_image
(
encoded_image_string_tensor
,
channels
=
3
)
image_tensor
.
set_shape
((
None
,
None
,
3
))
return
image_tensor
return
(
batch_image_str_placeholder
,
tf
.
map_fn
(
decode
,
elems
=
batch_image_str_placeholder
,
dtype
=
tf
.
uint8
,
parallel_iterations
=
32
,
back_prop
=
False
))
input_placeholder_fn_map
=
{
input_placeholder_fn_map
=
{
...
@@ -151,7 +173,8 @@ input_placeholder_fn_map = {
...
@@ -151,7 +173,8 @@ input_placeholder_fn_map = {
}
}
def
_add_output_tensor_nodes
(
postprocessed_tensors
):
def
_add_output_tensor_nodes
(
postprocessed_tensors
,
output_collection_name
=
'inference_op'
):
"""Adds output nodes for detection boxes and scores.
"""Adds output nodes for detection boxes and scores.
Adds the following nodes for output tensors -
Adds the following nodes for output tensors -
...
@@ -174,6 +197,7 @@ def _add_output_tensor_nodes(postprocessed_tensors):
...
@@ -174,6 +197,7 @@ def _add_output_tensor_nodes(postprocessed_tensors):
'detection_masks': [batch, max_detections, mask_height, mask_width]
'detection_masks': [batch, max_detections, mask_height, mask_width]
(optional).
(optional).
'num_detections': [batch]
'num_detections': [batch]
output_collection_name: Name of collection to add output tensors to.
Returns:
Returns:
A tensor dict containing the added output tensor nodes.
A tensor dict containing the added output tensor nodes.
...
@@ -191,53 +215,29 @@ def _add_output_tensor_nodes(postprocessed_tensors):
...
@@ -191,53 +215,29 @@ def _add_output_tensor_nodes(postprocessed_tensors):
outputs
[
'num_detections'
]
=
tf
.
identity
(
num_detections
,
name
=
'num_detections'
)
outputs
[
'num_detections'
]
=
tf
.
identity
(
num_detections
,
name
=
'num_detections'
)
if
masks
is
not
None
:
if
masks
is
not
None
:
outputs
[
'detection_masks'
]
=
tf
.
identity
(
masks
,
name
=
'detection_masks'
)
outputs
[
'detection_masks'
]
=
tf
.
identity
(
masks
,
name
=
'detection_masks'
)
for
output_key
in
outputs
:
tf
.
add_to_collection
(
output_collection_name
,
outputs
[
output_key
])
if
masks
is
not
None
:
tf
.
add_to_collection
(
output_collection_name
,
outputs
[
'detection_masks'
])
return
outputs
return
outputs
def
_write_inference_graph
(
inference_graph_path
,
def
_write_frozen_graph
(
frozen_graph_path
,
frozen_graph_def
):
checkpoint_path
=
None
,
"""Writes frozen graph to disk.
use_moving_averages
=
False
,
output_node_names
=
(
'num_detections,detection_scores,'
'detection_boxes,detection_classes'
)):
"""Writes inference graph to disk with the option to bake in weights.
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:
Args:
inference_graph_path: Path to write inference graph.
frozen_graph_path: Path to write inference graph.
checkpoint_path: Optional path to the checkpoint file.
frozen_graph_def: tf.GraphDef holding frozen graph.
use_moving_averages: Whether to export the original or the moving averages
of the trainable variables from the checkpoint.
output_node_names: Output tensor names, defaults are: num_detections,
detection_scores, detection_boxes, detection_classes.
"""
"""
inference_graph_def
=
tf
.
get_default_graph
().
as_graph_def
()
with
gfile
.
GFile
(
frozen_graph_path
,
'wb'
)
as
f
:
if
checkpoint_path
:
f
.
write
(
frozen_graph_def
.
SerializeToString
())
output_graph_def
=
get_frozen_graph_def
(
logging
.
info
(
'%d ops in the final graph.'
,
len
(
frozen_graph_def
.
node
))
inference_graph_def
=
inference_graph_def
,
use_moving_averages
=
use_moving_averages
,
input_checkpoint
=
checkpoint_path
,
def
_write_saved_model
(
saved_model_path
,
output_node_names
=
output_node_names
,
frozen_graph_def
,
)
inputs
,
outputs
):
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
),
os
.
path
.
basename
(
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.
"""Writes SavedModel to disk.
If checkpoint_path is not None bakes the weights into the graph thereby
If checkpoint_path is not None bakes the weights into the graph thereby
...
@@ -247,30 +247,17 @@ def _write_saved_model(inference_graph_path, inputs, outputs,
...
@@ -247,30 +247,17 @@ def _write_saved_model(inference_graph_path, inputs, outputs,
is restored.
is restored.
Args:
Args:
inference_graph_path: Path to write inference graph.
saved_model_path: Path to write SavedModel.
frozen_graph_def: tf.GraphDef holding frozen graph.
inputs: The input image tensor to use for detection.
inputs: The input image tensor to use for detection.
outputs: A tensor dictionary containing the outputs of a DetectionModel.
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
tf
.
Graph
().
as_default
():
with
session
.
Session
()
as
sess
:
with
session
.
Session
()
as
sess
:
tf
.
import_graph_def
(
checkpoint
_graph_def
)
tf
.
import_graph_def
(
frozen
_graph_def
,
name
=
''
)
builder
=
tf
.
saved_model
.
builder
.
SavedModelBuilder
(
inference_graph
_path
)
builder
=
tf
.
saved_model
.
builder
.
SavedModelBuilder
(
saved_model
_path
)
tensor_info_inputs
=
{
tensor_info_inputs
=
{
'inputs'
:
tf
.
saved_model
.
utils
.
build_tensor_info
(
inputs
)}
'inputs'
:
tf
.
saved_model
.
utils
.
build_tensor_info
(
inputs
)}
...
@@ -294,46 +281,96 @@ def _write_saved_model(inference_graph_path, inputs, outputs,
...
@@ -294,46 +281,96 @@ def _write_saved_model(inference_graph_path, inputs, outputs,
builder
.
save
()
builder
.
save
()
def
_write_graph_and_checkpoint
(
inference_graph_def
,
model_path
,
input_saver_def
,
trained_checkpoint_prefix
):
for
node
in
inference_graph_def
.
node
:
node
.
device
=
''
with
tf
.
Graph
().
as_default
():
tf
.
import_graph_def
(
inference_graph_def
,
name
=
''
)
with
session
.
Session
()
as
sess
:
saver
=
saver_lib
.
Saver
(
saver_def
=
input_saver_def
,
save_relative_paths
=
True
)
saver
.
restore
(
sess
,
trained_checkpoint_prefix
)
saver
.
save
(
sess
,
model_path
)
def
_export_inference_graph
(
input_type
,
def
_export_inference_graph
(
input_type
,
detection_model
,
detection_model
,
use_moving_averages
,
use_moving_averages
,
checkpoint_path
,
trained_checkpoint_prefix
,
inference_graph_path
,
output_directory
,
export_as_saved_model
=
False
):
optimize_graph
=
False
,
output_collection_name
=
'inference_op'
):
"""Export helper."""
"""Export helper."""
tf
.
gfile
.
MakeDirs
(
output_directory
)
frozen_graph_path
=
os
.
path
.
join
(
output_directory
,
'frozen_inference_graph.pb'
)
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
:
if
input_type
not
in
input_placeholder_fn_map
:
raise
ValueError
(
'Unknown input type: {}'
.
format
(
input_type
))
raise
ValueError
(
'Unknown input type: {}'
.
format
(
input_type
))
inputs
=
tf
.
to_float
(
input_placeholder_fn_map
[
input_type
]())
placeholder_tensor
,
input_tensors
=
input_placeholder_fn_map
[
input_type
]()
inputs
=
tf
.
to_float
(
input_tensors
)
preprocessed_inputs
=
detection_model
.
preprocess
(
inputs
)
preprocessed_inputs
=
detection_model
.
preprocess
(
inputs
)
output_tensors
=
detection_model
.
predict
(
preprocessed_inputs
)
output_tensors
=
detection_model
.
predict
(
preprocessed_inputs
)
postprocessed_tensors
=
detection_model
.
postprocess
(
output_tensors
)
postprocessed_tensors
=
detection_model
.
postprocess
(
output_tensors
)
outputs
=
_add_output_tensor_nodes
(
postprocessed_tensors
)
outputs
=
_add_output_tensor_nodes
(
postprocessed_tensors
,
out_node_names
=
list
(
outputs
.
keys
())
output_collection_name
)
if
export_as_saved_model
:
_write_saved_model
(
inference_graph_path
,
inputs
,
outputs
,
checkpoint_path
,
saver
=
None
use_moving_averages
)
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
:
else
:
_write_inference_graph
(
inference_graph_path
,
checkpoint_path
,
saver
=
tf
.
train
.
Saver
()
use_moving_averages
,
input_saver_def
=
saver
.
as_saver_def
()
output_node_names
=
','
.
join
(
out_node_names
))
_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
)
frozen_graph_def
=
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
=
','
.
join
(
outputs
.
keys
()),
restore_op_name
=
'save/restore_all'
,
filename_tensor_name
=
'save/Const:0'
,
clear_devices
=
True
,
optimize_graph
=
optimize_graph
,
initializer_nodes
=
''
)
_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
,
pipeline_config
,
checkpoint_path
,
def
export_inference_graph
(
input_type
,
inference_graph_path
,
export_as_saved_model
=
False
):
pipeline_config
,
trained_checkpoint_prefix
,
output_directory
,
optimize_graph
=
False
,
output_collection_name
=
'inference_op'
):
"""Exports inference graph for the model specified in the pipeline config.
"""Exports inference graph for the model specified in the pipeline config.
Args:
Args:
input_type: Type of input for the graph. Can be one of [`image_tensor`,
input_type: Type of input for the graph. Can be one of [`image_tensor`,
`tf_example`].
`tf_example`].
pipeline_config: pipeline_pb2.TrainAndEvalPipelineConfig proto.
pipeline_config: pipeline_pb2.TrainAndEvalPipelineConfig proto.
checkpoint_path: Path to the checkpoint file to freeze.
trained_checkpoint_prefix: Path to the trained checkpoint file.
inference_graph_path: Path to write inference graph to.
output_directory: Path to write outputs.
export_as_saved_model: If the model should be exported as a SavedModel. If
optimize_graph: Whether to optimize graph using Grappler.
false, it is saved as an inference graph.
output_collection_name: Name of collection to add output tensors to.
If None, does not add output tensors to a collection.
"""
"""
detection_model
=
model_builder
.
build
(
pipeline_config
.
model
,
detection_model
=
model_builder
.
build
(
pipeline_config
.
model
,
is_training
=
False
)
is_training
=
False
)
_export_inference_graph
(
input_type
,
detection_model
,
_export_inference_graph
(
input_type
,
detection_model
,
pipeline_config
.
eval_config
.
use_moving_averages
,
pipeline_config
.
eval_config
.
use_moving_averages
,
checkpoint_p
ath
,
inference_graph_path
,
trained_
checkpoint_p
refix
,
output_directory
,
export_as_saved_model
)
optimize_graph
,
output_collection_name
)
object_detection/exporter_test.py
View file @
582bf927
This diff is collapsed.
Click to expand it.
object_detection/meta_architectures/BUILD
View file @
582bf927
...
@@ -13,12 +13,11 @@ py_library(
...
@@ -13,12 +13,11 @@ py_library(
srcs
=
[
"ssd_meta_arch.py"
],
srcs
=
[
"ssd_meta_arch.py"
],
deps
=
[
deps
=
[
"//tensorflow"
,
"//tensorflow"
,
"//tensorflow_models/object_detection/core:box_coder"
,
"//tensorflow_models/object_detection/core:box_list"
,
"//tensorflow_models/object_detection/core:box_list"
,
"//tensorflow_models/object_detection/core:box_predictor"
,
"//tensorflow_models/object_detection/core:box_predictor"
,
"//tensorflow_models/object_detection/core:model"
,
"//tensorflow_models/object_detection/core:model"
,
"//tensorflow_models/object_detection/core:target_assigner"
,
"//tensorflow_models/object_detection/core:target_assigner"
,
"//tensorflow_models/object_detection/utils:
variables_helper
"
,
"//tensorflow_models/object_detection/utils:
shape_utils
"
,
],
],
)
)
...
@@ -56,7 +55,7 @@ py_library(
...
@@ -56,7 +55,7 @@ py_library(
"//tensorflow_models/object_detection/core:standard_fields"
,
"//tensorflow_models/object_detection/core:standard_fields"
,
"//tensorflow_models/object_detection/core:target_assigner"
,
"//tensorflow_models/object_detection/core:target_assigner"
,
"//tensorflow_models/object_detection/utils:ops"
,
"//tensorflow_models/object_detection/utils:ops"
,
"//tensorflow_models/object_detection/utils:
variables_helper
"
,
"//tensorflow_models/object_detection/utils:
shape_utils
"
,
],
],
)
)
...
...
object_detection/meta_architectures/faster_rcnn_meta_arch.py
View file @
582bf927
...
@@ -80,7 +80,7 @@ from object_detection.core import post_processing
...
@@ -80,7 +80,7 @@ from object_detection.core import post_processing
from
object_detection.core
import
standard_fields
as
fields
from
object_detection.core
import
standard_fields
as
fields
from
object_detection.core
import
target_assigner
from
object_detection.core
import
target_assigner
from
object_detection.utils
import
ops
from
object_detection.utils
import
ops
from
object_detection.utils
import
variables_helper
from
object_detection.utils
import
shape_utils
slim
=
tf
.
contrib
.
slim
slim
=
tf
.
contrib
.
slim
...
@@ -159,21 +159,19 @@ class FasterRCNNFeatureExtractor(object):
...
@@ -159,21 +159,19 @@ class FasterRCNNFeatureExtractor(object):
def
restore_from_classification_checkpoint_fn
(
def
restore_from_classification_checkpoint_fn
(
self
,
self
,
checkpoint_path
,
first_stage_feature_extractor_scope
,
first_stage_feature_extractor_scope
,
second_stage_feature_extractor_scope
):
second_stage_feature_extractor_scope
):
"""Returns
callable for loading a checkpoint into the tensorflow graph
.
"""Returns
a map of variables to load from a foreign checkpoint
.
Args:
Args:
checkpoint_path: path to checkpoint to restore.
first_stage_feature_extractor_scope: A scope name for the first stage
first_stage_feature_extractor_scope: A scope name for the first stage
feature extractor.
feature extractor.
second_stage_feature_extractor_scope: A scope name for the second stage
second_stage_feature_extractor_scope: A scope name for the second stage
feature extractor.
feature extractor.
Returns:
Returns:
a callable which takes a tf.Session as input and loads a checkpoint whe
n
A dict mapping variable names (to load from a checkpoint) to variables i
n
run
.
the model graph
.
"""
"""
variables_to_restore
=
{}
variables_to_restore
=
{}
for
variable
in
tf
.
global_variables
():
for
variable
in
tf
.
global_variables
():
...
@@ -182,13 +180,7 @@ class FasterRCNNFeatureExtractor(object):
...
@@ -182,13 +180,7 @@ class FasterRCNNFeatureExtractor(object):
if
variable
.
op
.
name
.
startswith
(
scope_name
):
if
variable
.
op
.
name
.
startswith
(
scope_name
):
var_name
=
variable
.
op
.
name
.
replace
(
scope_name
+
'/'
,
''
)
var_name
=
variable
.
op
.
name
.
replace
(
scope_name
+
'/'
,
''
)
variables_to_restore
[
var_name
]
=
variable
variables_to_restore
[
var_name
]
=
variable
variables_to_restore
=
(
return
variables_to_restore
variables_helper
.
get_variables_available_in_checkpoint
(
variables_to_restore
,
checkpoint_path
))
saver
=
tf
.
train
.
Saver
(
variables_to_restore
)
def
restore
(
sess
):
saver
.
restore
(
sess
,
checkpoint_path
)
return
restore
class
FasterRCNNMetaArch
(
model
.
DetectionModel
):
class
FasterRCNNMetaArch
(
model
.
DetectionModel
):
...
@@ -774,10 +766,9 @@ class FasterRCNNMetaArch(model.DetectionModel):
...
@@ -774,10 +766,9 @@ class FasterRCNNMetaArch(model.DetectionModel):
A float tensor with shape [A * B, ..., depth] (where the first and last
A float tensor with shape [A * B, ..., depth] (where the first and last
dimension are statically defined.
dimension are statically defined.
"""
"""
inputs_shape
=
inputs
.
get_shape
().
as_list
()
combined_shape
=
shape_utils
.
combined_static_and_dynamic_shape
(
inputs
)
flattened_shape
=
tf
.
concat
([
flattened_shape
=
tf
.
stack
([
combined_shape
[
0
]
*
combined_shape
[
1
]]
+
[
inputs_shape
[
0
]
*
inputs_shape
[
1
]],
tf
.
shape
(
inputs
)[
2
:
-
1
],
combined_shape
[
2
:])
[
inputs_shape
[
-
1
]]],
0
)
return
tf
.
reshape
(
inputs
,
flattened_shape
)
return
tf
.
reshape
(
inputs
,
flattened_shape
)
def
postprocess
(
self
,
prediction_dict
):
def
postprocess
(
self
,
prediction_dict
):
...
@@ -875,52 +866,128 @@ class FasterRCNNMetaArch(model.DetectionModel):
...
@@ -875,52 +866,128 @@ class FasterRCNNMetaArch(model.DetectionModel):
representing the number of proposals predicted for each image in
representing the number of proposals predicted for each image in
the batch.
the batch.
"""
"""
rpn_box_encodings_batch
=
tf
.
expand_dims
(
rpn_box_encodings_batch
,
axis
=
2
)
rpn_encodings_shape
=
shape_utils
.
combined_static_and_dynamic_shape
(
rpn_box_encodings_batch
)
tiled_anchor_boxes
=
tf
.
tile
(
tf
.
expand_dims
(
anchors
,
0
),
[
rpn_encodings_shape
[
0
],
1
,
1
])
proposal_boxes
=
self
.
_batch_decode_boxes
(
rpn_box_encodings_batch
,
tiled_anchor_boxes
)
proposal_boxes
=
tf
.
squeeze
(
proposal_boxes
,
axis
=
2
)
rpn_objectness_softmax_without_background
=
tf
.
nn
.
softmax
(
rpn_objectness_predictions_with_background_batch
)[:,
:,
1
]
clip_window
=
tf
.
to_float
(
tf
.
stack
([
0
,
0
,
image_shape
[
1
],
image_shape
[
2
]]))
clip_window
=
tf
.
to_float
(
tf
.
stack
([
0
,
0
,
image_shape
[
1
],
image_shape
[
2
]]))
(
proposal_boxes
,
proposal_scores
,
_
,
_
,
num_proposals
)
=
post_processing
.
batch_multiclass_non_max_suppression
(
tf
.
expand_dims
(
proposal_boxes
,
axis
=
2
),
tf
.
expand_dims
(
rpn_objectness_softmax_without_background
,
axis
=
2
),
self
.
_first_stage_nms_score_threshold
,
self
.
_first_stage_nms_iou_threshold
,
self
.
_first_stage_max_proposals
,
self
.
_first_stage_max_proposals
,
clip_window
=
clip_window
)
if
self
.
_is_training
:
if
self
.
_is_training
:
(
groundtruth_boxlists
,
groundtruth_classes_with_background_list
proposal_boxes
=
tf
.
stop_gradient
(
proposal_boxes
)
)
=
self
.
_format_groundtruth_data
(
image_shape
)
if
not
self
.
_hard_example_miner
:
(
groundtruth_boxlists
,
groundtruth_classes_with_background_list
,
proposal_boxes_list
=
[]
)
=
self
.
_format_groundtruth_data
(
image_shape
)
proposal_scores_list
=
[]
(
proposal_boxes
,
proposal_scores
,
num_proposals_list
=
[]
num_proposals
)
=
self
.
_unpad_proposals_and_sample_box_classifier_batch
(
for
(
batch_index
,
proposal_boxes
,
proposal_scores
,
num_proposals
,
(
rpn_box_encodings
,
groundtruth_boxlists
,
groundtruth_classes_with_background_list
)
rpn_objectness_predictions_with_background
))
in
enumerate
(
zip
(
# normalize proposal boxes
tf
.
unstack
(
rpn_box_encodings_batch
),
proposal_boxes_reshaped
=
tf
.
reshape
(
proposal_boxes
,
[
-
1
,
4
])
tf
.
unstack
(
rpn_objectness_predictions_with_background_batch
))):
normalized_proposal_boxes_reshaped
=
box_list_ops
.
to_normalized_coordinates
(
decoded_boxes
=
self
.
_box_coder
.
decode
(
box_list
.
BoxList
(
proposal_boxes_reshaped
),
rpn_box_encodings
,
box_list
.
BoxList
(
anchors
))
image_shape
[
1
],
image_shape
[
2
],
check_range
=
False
).
get
()
objectness_scores
=
tf
.
unstack
(
proposal_boxes
=
tf
.
reshape
(
normalized_proposal_boxes_reshaped
,
tf
.
nn
.
softmax
(
rpn_objectness_predictions_with_background
),
axis
=
1
)[
1
]
[
-
1
,
proposal_boxes
.
shape
[
1
].
value
,
4
])
proposal_boxlist
=
post_processing
.
multiclass_non_max_suppression
(
return
proposal_boxes
,
proposal_scores
,
num_proposals
tf
.
expand_dims
(
decoded_boxes
.
get
(),
1
),
tf
.
expand_dims
(
objectness_scores
,
1
),
def
_unpad_proposals_and_sample_box_classifier_batch
(
self
.
_first_stage_nms_score_threshold
,
self
,
self
.
_first_stage_nms_iou_threshold
,
self
.
_first_stage_max_proposals
,
proposal_boxes
,
clip_window
=
clip_window
)
proposal_scores
,
num_proposals
,
if
self
.
_is_training
:
groundtruth_boxlists
,
proposal_boxlist
.
set
(
tf
.
stop_gradient
(
proposal_boxlist
.
get
()))
groundtruth_classes_with_background_list
):
if
not
self
.
_hard_example_miner
:
"""Unpads proposals and samples a minibatch for second stage.
proposal_boxlist
=
self
.
_sample_box_classifier_minibatch
(
proposal_boxlist
,
groundtruth_boxlists
[
batch_index
],
Args:
groundtruth_classes_with_background_list
[
batch_index
])
proposal_boxes: A float tensor with shape
[batch_size, num_proposals, 4] representing the (potentially zero
normalized_proposals
=
box_list_ops
.
to_normalized_coordinates
(
padded) proposal boxes for all images in the batch. These boxes are
proposal_boxlist
,
image_shape
[
1
],
image_shape
[
2
],
represented as normalized coordinates.
check_range
=
False
)
proposal_scores: A float tensor with shape
[batch_size, num_proposals] representing the (potentially zero
# pad proposals to max_num_proposals
padded) proposal objectness scores for all images in the batch.
padded_proposals
=
box_list_ops
.
pad_or_clip_box_list
(
num_proposals: A Tensor of type `int32`. A 1-D tensor of shape [batch]
normalized_proposals
,
num_boxes
=
self
.
max_num_proposals
)
representing the number of proposals predicted for each image in
proposal_boxes_list
.
append
(
padded_proposals
.
get
())
the batch.
proposal_scores_list
.
append
(
groundtruth_boxlists: A list of BoxLists containing (absolute) coordinates
padded_proposals
.
get_field
(
fields
.
BoxListFields
.
scores
))
of the groundtruth boxes.
num_proposals_list
.
append
(
tf
.
minimum
(
normalized_proposals
.
num_boxes
(),
groundtruth_classes_with_background_list: A list of 2-D one-hot
self
.
max_num_proposals
))
(or k-hot) tensors of shape [num_boxes, num_classes+1] containing the
class targets with the 0th index assumed to map to the background class.
return
(
tf
.
stack
(
proposal_boxes_list
),
tf
.
stack
(
proposal_scores_list
),
tf
.
stack
(
num_proposals_list
))
Returns:
proposal_boxes: A float tensor with shape
[batch_size, second_stage_batch_size, 4] representing the (potentially
zero padded) proposal boxes for all images in the batch. These boxes
are represented as normalized coordinates.
proposal_scores: A float tensor with shape
[batch_size, second_stage_batch_size] representing the (potentially zero
padded) proposal objectness scores for all images in the batch.
num_proposals: A Tensor of type `int32`. A 1-D tensor of shape [batch]
representing the number of proposals predicted for each image in
the batch.
"""
single_image_proposal_box_sample
=
[]
single_image_proposal_score_sample
=
[]
single_image_num_proposals_sample
=
[]
for
(
single_image_proposal_boxes
,
single_image_proposal_scores
,
single_image_num_proposals
,
single_image_groundtruth_boxlist
,
single_image_groundtruth_classes_with_background
)
in
zip
(
tf
.
unstack
(
proposal_boxes
),
tf
.
unstack
(
proposal_scores
),
tf
.
unstack
(
num_proposals
),
groundtruth_boxlists
,
groundtruth_classes_with_background_list
):
static_shape
=
single_image_proposal_boxes
.
get_shape
()
sliced_static_shape
=
tf
.
TensorShape
([
tf
.
Dimension
(
None
),
static_shape
.
dims
[
-
1
]])
single_image_proposal_boxes
=
tf
.
slice
(
single_image_proposal_boxes
,
[
0
,
0
],
[
single_image_num_proposals
,
-
1
])
single_image_proposal_boxes
.
set_shape
(
sliced_static_shape
)
single_image_proposal_scores
=
tf
.
slice
(
single_image_proposal_scores
,
[
0
],
[
single_image_num_proposals
])
single_image_boxlist
=
box_list
.
BoxList
(
single_image_proposal_boxes
)
single_image_boxlist
.
add_field
(
fields
.
BoxListFields
.
scores
,
single_image_proposal_scores
)
sampled_boxlist
=
self
.
_sample_box_classifier_minibatch
(
single_image_boxlist
,
single_image_groundtruth_boxlist
,
single_image_groundtruth_classes_with_background
)
sampled_padded_boxlist
=
box_list_ops
.
pad_or_clip_box_list
(
sampled_boxlist
,
num_boxes
=
self
.
_second_stage_batch_size
)
single_image_num_proposals_sample
.
append
(
tf
.
minimum
(
sampled_boxlist
.
num_boxes
(),
self
.
_second_stage_batch_size
))
bb
=
sampled_padded_boxlist
.
get
()
single_image_proposal_box_sample
.
append
(
bb
)
single_image_proposal_score_sample
.
append
(
sampled_padded_boxlist
.
get_field
(
fields
.
BoxListFields
.
scores
))
return
(
tf
.
stack
(
single_image_proposal_box_sample
),
tf
.
stack
(
single_image_proposal_score_sample
),
tf
.
stack
(
single_image_num_proposals_sample
))
def
_format_groundtruth_data
(
self
,
image_shape
):
def
_format_groundtruth_data
(
self
,
image_shape
):
"""Helper function for preparing groundtruth data for target assignment.
"""Helper function for preparing groundtruth data for target assignment.
...
@@ -1074,7 +1141,7 @@ class FasterRCNNMetaArch(model.DetectionModel):
...
@@ -1074,7 +1141,7 @@ class FasterRCNNMetaArch(model.DetectionModel):
class_predictions_with_background
,
class_predictions_with_background
,
[
-
1
,
self
.
max_num_proposals
,
self
.
num_classes
+
1
]
[
-
1
,
self
.
max_num_proposals
,
self
.
num_classes
+
1
]
)
)
refined_decoded_boxes_batch
=
self
.
_batch_decode_
refined_
boxes
(
refined_decoded_boxes_batch
=
self
.
_batch_decode_boxes
(
refined_box_encodings_batch
,
proposal_boxes
)
refined_box_encodings_batch
,
proposal_boxes
)
class_predictions_with_background_batch
=
(
class_predictions_with_background_batch
=
(
self
.
_second_stage_score_conversion_fn
(
self
.
_second_stage_score_conversion_fn
(
...
@@ -1092,19 +1159,26 @@ class FasterRCNNMetaArch(model.DetectionModel):
...
@@ -1092,19 +1159,26 @@ class FasterRCNNMetaArch(model.DetectionModel):
mask_predictions_batch
=
tf
.
reshape
(
mask_predictions_batch
=
tf
.
reshape
(
mask_predictions
,
[
-
1
,
self
.
max_num_proposals
,
mask_predictions
,
[
-
1
,
self
.
max_num_proposals
,
self
.
num_classes
,
mask_height
,
mask_width
])
self
.
num_classes
,
mask_height
,
mask_width
])
detections
=
self
.
_second_stage_nms_fn
(
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
nmsed_masks
,
refined_decoded_boxes_batch
,
num_detections
)
=
self
.
_second_stage_nms_fn
(
class_predictions_batch
,
refined_decoded_boxes_batch
,
clip_window
=
clip_window
,
class_predictions_batch
,
change_coordinate_frame
=
True
,
clip_window
=
clip_window
,
num_valid_boxes
=
num_proposals
,
change_coordinate_frame
=
True
,
masks
=
mask_predictions_batch
)
num_valid_boxes
=
num_proposals
,
masks
=
mask_predictions_batch
)
detections
=
{
'detection_boxes'
:
nmsed_boxes
,
'detection_scores'
:
nmsed_scores
,
'detection_classes'
:
nmsed_classes
,
'num_detections'
:
tf
.
to_float
(
num_detections
)}
if
nmsed_masks
is
not
None
:
detections
[
'detection_masks'
]
=
nmsed_masks
if
mask_predictions
is
not
None
:
if
mask_predictions
is
not
None
:
detections
[
'detection_masks'
]
=
tf
.
to_float
(
detections
[
'detection_masks'
]
=
tf
.
to_float
(
tf
.
greater_equal
(
detections
[
'detection_masks'
],
mask_threshold
))
tf
.
greater_equal
(
detections
[
'detection_masks'
],
mask_threshold
))
return
detections
return
detections
def
_batch_decode_
refined_
boxes
(
self
,
refined_
box_encodings
,
proposal
_boxes
):
def
_batch_decode_boxes
(
self
,
box_encodings
,
anchor
_boxes
):
"""Decode tensor of refined box encodings.
"""Decode tensor of refined box encodings.
Args:
Args:
...
@@ -1119,15 +1193,33 @@ class FasterRCNNMetaArch(model.DetectionModel):
...
@@ -1119,15 +1193,33 @@ class FasterRCNNMetaArch(model.DetectionModel):
float tensor representing (padded) refined bounding box predictions
float tensor representing (padded) refined bounding box predictions
(for each image in batch, proposal and class).
(for each image in batch, proposal and class).
"""
"""
tiled_proposal_boxes
=
tf
.
tile
(
"""Decodes box encodings with respect to the anchor boxes.
tf
.
expand_dims
(
proposal_boxes
,
2
),
[
1
,
1
,
self
.
num_classes
,
1
])
tiled_proposals_boxlist
=
box_list
.
BoxList
(
Args:
tf
.
reshape
(
tiled_proposal_boxes
,
[
-
1
,
4
]))
box_encodings: a 4-D tensor with shape
[batch_size, num_anchors, num_classes, self._box_coder.code_size]
representing box encodings.
anchor_boxes: [batch_size, num_anchors, 4] representing
decoded bounding boxes.
Returns:
decoded_boxes: a [batch_size, num_anchors, num_classes, 4]
float tensor representing bounding box predictions
(for each image in batch, proposal and class).
"""
combined_shape
=
shape_utils
.
combined_static_and_dynamic_shape
(
box_encodings
)
num_classes
=
combined_shape
[
2
]
tiled_anchor_boxes
=
tf
.
tile
(
tf
.
expand_dims
(
anchor_boxes
,
2
),
[
1
,
1
,
num_classes
,
1
])
tiled_anchors_boxlist
=
box_list
.
BoxList
(
tf
.
reshape
(
tiled_anchor_boxes
,
[
-
1
,
4
]))
decoded_boxes
=
self
.
_box_coder
.
decode
(
decoded_boxes
=
self
.
_box_coder
.
decode
(
tf
.
reshape
(
refined_
box_encodings
,
[
-
1
,
self
.
_box_coder
.
code_size
]),
tf
.
reshape
(
box_encodings
,
[
-
1
,
self
.
_box_coder
.
code_size
]),
tiled_
proposal
s_boxlist
)
tiled_
anchor
s_boxlist
)
return
tf
.
reshape
(
decoded_boxes
.
get
(),
return
tf
.
reshape
(
decoded_boxes
.
get
(),
[
-
1
,
self
.
max_num_proposals
,
self
.
num_classes
,
4
])
tf
.
stack
([
combined_shape
[
0
],
combined_shape
[
1
],
num_classes
,
4
]))
def
loss
(
self
,
prediction_dict
,
scope
=
None
):
def
loss
(
self
,
prediction_dict
,
scope
=
None
):
"""Compute scalar loss tensors given prediction tensors.
"""Compute scalar loss tensors given prediction tensors.
...
@@ -1413,25 +1505,22 @@ class FasterRCNNMetaArch(model.DetectionModel):
...
@@ -1413,25 +1505,22 @@ class FasterRCNNMetaArch(model.DetectionModel):
cls_losses
=
tf
.
expand_dims
(
single_image_cls_loss
,
0
),
cls_losses
=
tf
.
expand_dims
(
single_image_cls_loss
,
0
),
decoded_boxlist_list
=
[
proposal_boxlist
])
decoded_boxlist_list
=
[
proposal_boxlist
])
def
restore_fn
(
self
,
checkpoint_path
,
from_detection_checkpoint
=
True
):
def
restore_map
(
self
,
from_detection_checkpoint
=
True
):
"""Returns callable for loading a checkpoint into the tensorflow graph.
"""Returns a map of variables to load from a foreign checkpoint.
See parent class for details.
Args:
Args:
checkpoint_path: path to checkpoint to restore.
from_detection_checkpoint: whether to restore from a full detection
from_detection_checkpoint: whether to restore from a detection checkpoint
checkpoint (with compatible variable names) or to restore from a
(with compatible variable names) or to restore from a classification
classification checkpoint for initialization prior to training.
checkpoint for initialization prior to training. Note that when
from_detection_checkpoint=True, the current implementation only
supports restoration from an (exactly) identical model (with exception
of the num_classes parameter).
Returns:
Returns:
a callable which takes a tf.Session as input and loads a checkpoint whe
n
A dict mapping variable names (to load from a checkpoint) to variables i
n
run
.
the model graph
.
"""
"""
if
not
from_detection_checkpoint
:
if
not
from_detection_checkpoint
:
return
self
.
_feature_extractor
.
restore_from_classification_checkpoint_fn
(
return
self
.
_feature_extractor
.
restore_from_classification_checkpoint_fn
(
checkpoint_path
,
self
.
first_stage_feature_extractor_scope
,
self
.
first_stage_feature_extractor_scope
,
self
.
second_stage_feature_extractor_scope
)
self
.
second_stage_feature_extractor_scope
)
...
@@ -1439,13 +1528,8 @@ class FasterRCNNMetaArch(model.DetectionModel):
...
@@ -1439,13 +1528,8 @@ class FasterRCNNMetaArch(model.DetectionModel):
variables_to_restore
.
append
(
slim
.
get_or_create_global_step
())
variables_to_restore
.
append
(
slim
.
get_or_create_global_step
())
# Only load feature extractor variables to be consistent with loading from
# Only load feature extractor variables to be consistent with loading from
# a classification checkpoint.
# a classification checkpoint.
f
irst_stage
_variables
=
tf
.
contrib
.
framework
.
filter_variables
(
f
eature_extractor
_variables
=
tf
.
contrib
.
framework
.
filter_variables
(
variables_to_restore
,
variables_to_restore
,
include_patterns
=
[
self
.
first_stage_feature_extractor_scope
,
include_patterns
=
[
self
.
first_stage_feature_extractor_scope
,
self
.
second_stage_feature_extractor_scope
])
self
.
second_stage_feature_extractor_scope
])
return
{
var
.
op
.
name
:
var
for
var
in
feature_extractor_variables
}
saver
=
tf
.
train
.
Saver
(
first_stage_variables
)
def
restore
(
sess
):
saver
.
restore
(
sess
,
checkpoint_path
)
return
restore
object_detection/meta_architectures/faster_rcnn_meta_arch_test_lib.py
View file @
582bf927
...
@@ -226,61 +226,47 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
...
@@ -226,61 +226,47 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
return
self
.
_get_model
(
self
.
_get_second_stage_box_predictor
(
return
self
.
_get_model
(
self
.
_get_second_stage_box_predictor
(
num_classes
=
num_classes
,
is_training
=
is_training
),
**
common_kwargs
)
num_classes
=
num_classes
,
is_training
=
is_training
),
**
common_kwargs
)
def
test_predict_
gives_
correct_shapes_in_inference_mode_
first
_stage
_only
(
def
test_predict_correct_shapes_in_inference_mode_
both
_stage
s
(
self
):
self
):
test_graph
=
tf
.
Graph
()
batch_size
=
2
with
test_graph
.
as_default
():
image_size
=
10
model
=
self
.
_build_model
(
input_shapes
=
[(
batch_size
,
image_size
,
image_size
,
3
),
is_training
=
False
,
first_stage_only
=
True
,
second_stage_batch_size
=
2
)
(
None
,
image_size
,
image_size
,
3
),
batch_size
=
2
(
batch_size
,
None
,
None
,
3
),
height
=
10
(
None
,
None
,
None
,
3
)]
width
=
12
expected_num_anchors
=
image_size
*
image_size
*
3
*
3
input_image_shape
=
(
batch_size
,
height
,
width
,
3
)
expected_shapes
=
{
'rpn_box_predictor_features'
:
preprocessed_inputs
=
tf
.
placeholder
(
dtype
=
tf
.
float32
,
(
2
,
image_size
,
image_size
,
512
),
shape
=
(
batch_size
,
None
,
None
,
3
))
'rpn_features_to_crop'
:
(
2
,
image_size
,
image_size
,
3
),
prediction_dict
=
model
.
predict
(
preprocessed_inputs
)
'image_shape'
:
(
4
,),
'rpn_box_encodings'
:
(
2
,
expected_num_anchors
,
4
),
# In inference mode, anchors are clipped to the image window, but not
'rpn_objectness_predictions_with_background'
:
# pruned. Since MockFasterRCNN.extract_proposal_features returns a
(
2
,
expected_num_anchors
,
2
),
# tensor with the same shape as its input, the expected number of anchors
'anchors'
:
(
expected_num_anchors
,
4
),
# is height * width * the number of anchors per location (i.e. 3x3).
'refined_box_encodings'
:
(
2
*
8
,
2
,
4
),
expected_num_anchors
=
height
*
width
*
3
*
3
'class_predictions_with_background'
:
(
2
*
8
,
2
+
1
),
expected_output_keys
=
set
([
'num_proposals'
:
(
2
,),
'rpn_box_predictor_features'
,
'rpn_features_to_crop'
,
'image_shape'
,
'proposal_boxes'
:
(
2
,
8
,
4
),
'rpn_box_encodings'
,
'rpn_objectness_predictions_with_background'
,
}
'anchors'
])
for
input_shape
in
input_shapes
:
expected_output_shapes
=
{
test_graph
=
tf
.
Graph
()
'rpn_box_predictor_features'
:
(
batch_size
,
height
,
width
,
512
),
with
test_graph
.
as_default
():
'rpn_features_to_crop'
:
(
batch_size
,
height
,
width
,
3
),
model
=
self
.
_build_model
(
'rpn_box_encodings'
:
(
batch_size
,
expected_num_anchors
,
4
),
is_training
=
False
,
first_stage_only
=
False
,
'rpn_objectness_predictions_with_background'
:
second_stage_batch_size
=
2
)
(
batch_size
,
expected_num_anchors
,
2
),
preprocessed_inputs
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
input_shape
)
'anchors'
:
(
expected_num_anchors
,
4
)
result_tensor_dict
=
model
.
predict
(
preprocessed_inputs
)
}
init_op
=
tf
.
global_variables_initializer
()
with
self
.
test_session
(
graph
=
test_graph
)
as
sess
:
init_op
=
tf
.
global_variables_initializer
()
with
self
.
test_session
()
as
sess
:
sess
.
run
(
init_op
)
sess
.
run
(
init_op
)
prediction_out
=
sess
.
run
(
prediction_dict
,
tensor_dict_out
=
sess
.
run
(
result_tensor_dict
,
feed_dict
=
{
feed_dict
=
{
preprocessed_inputs
:
preprocessed_inputs
:
np
.
zeros
((
batch_size
,
image_size
,
image_size
,
3
))})
np
.
zeros
(
input_image_shape
)
self
.
assertEqual
(
set
(
tensor_dict_out
.
keys
()),
})
set
(
expected_shapes
.
keys
()))
for
key
in
expected_shapes
:
self
.
assertEqual
(
set
(
prediction_out
.
keys
()),
expected_output_keys
)
self
.
assertAllEqual
(
tensor_dict_out
[
key
].
shape
,
expected_shapes
[
key
])
self
.
assertAllEqual
(
prediction_out
[
'image_shape'
],
input_image_shape
)
for
output_key
,
expected_shape
in
expected_output_shapes
.
items
():
self
.
assertAllEqual
(
prediction_out
[
output_key
].
shape
,
expected_shape
)
# Check that anchors are clipped to window.
anchors
=
prediction_out
[
'anchors'
]
self
.
assertTrue
(
np
.
all
(
np
.
greater_equal
(
anchors
,
0
)))
self
.
assertTrue
(
np
.
all
(
np
.
less_equal
(
anchors
[:,
0
],
height
)))
self
.
assertTrue
(
np
.
all
(
np
.
less_equal
(
anchors
[:,
1
],
width
)))
self
.
assertTrue
(
np
.
all
(
np
.
less_equal
(
anchors
[:,
2
],
height
)))
self
.
assertTrue
(
np
.
all
(
np
.
less_equal
(
anchors
[:,
3
],
width
)))
def
test_predict_gives_valid_anchors_in_training_mode_first_stage_only
(
self
):
def
test_predict_gives_valid_anchors_in_training_mode_first_stage_only
(
self
):
test_graph
=
tf
.
Graph
()
test_graph
=
tf
.
Graph
()
...
@@ -535,35 +521,67 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
...
@@ -535,35 +521,67 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
expected_num_proposals
)
expected_num_proposals
)
def
test_postprocess_second_stage_only_inference_mode
(
self
):
def
test_postprocess_second_stage_only_inference_mode
(
self
):
model
=
self
.
_build_model
(
num_proposals_shapes
=
[(
2
),
(
None
)]
is_training
=
False
,
first_stage_only
=
False
,
second_stage_batch_size
=
6
)
refined_box_encodings_shapes
=
[(
16
,
2
,
4
),
(
None
,
2
,
4
)]
class_predictions_with_background_shapes
=
[(
16
,
3
),
(
None
,
3
)]
proposal_boxes_shapes
=
[(
2
,
8
,
4
),
(
None
,
8
,
4
)]
batch_size
=
2
batch_size
=
2
total_num_padded_proposals
=
batch_size
*
model
.
max_num_proposals
image_shape
=
np
.
array
((
2
,
36
,
48
,
3
),
dtype
=
np
.
int32
)
proposal_boxes
=
tf
.
constant
(
for
(
num_proposals_shape
,
refined_box_encoding_shape
,
[[[
1
,
1
,
2
,
3
],
class_predictions_with_background_shape
,
[
0
,
0
,
1
,
1
],
proposal_boxes_shape
)
in
zip
(
num_proposals_shapes
,
[.
5
,
.
5
,
.
6
,
.
6
],
refined_box_encodings_shapes
,
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
]],
class_predictions_with_background_shapes
,
[[
2
,
3
,
6
,
8
],
proposal_boxes_shapes
):
[
1
,
2
,
5
,
3
],
tf_graph
=
tf
.
Graph
()
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
]]],
dtype
=
tf
.
float32
)
with
tf_graph
.
as_default
():
num_proposals
=
tf
.
constant
([
3
,
2
],
dtype
=
tf
.
int32
)
model
=
self
.
_build_model
(
refined_box_encodings
=
tf
.
zeros
(
is_training
=
False
,
first_stage_only
=
False
,
[
total_num_padded_proposals
,
model
.
num_classes
,
4
],
dtype
=
tf
.
float32
)
second_stage_batch_size
=
6
)
class_predictions_with_background
=
tf
.
ones
(
total_num_padded_proposals
=
batch_size
*
model
.
max_num_proposals
[
total_num_padded_proposals
,
model
.
num_classes
+
1
],
dtype
=
tf
.
float32
)
proposal_boxes
=
np
.
array
(
image_shape
=
tf
.
constant
([
batch_size
,
36
,
48
,
3
],
dtype
=
tf
.
int32
)
[[[
1
,
1
,
2
,
3
],
[
0
,
0
,
1
,
1
],
detections
=
model
.
postprocess
({
[.
5
,
.
5
,
.
6
,
.
6
],
'refined_box_encodings'
:
refined_box_encodings
,
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
]],
'class_predictions_with_background'
:
class_predictions_with_background
,
[[
2
,
3
,
6
,
8
],
'num_proposals'
:
num_proposals
,
[
1
,
2
,
5
,
3
],
'proposal_boxes'
:
proposal_boxes
,
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
],
4
*
[
0
]]])
'image_shape'
:
image_shape
num_proposals
=
np
.
array
([
3
,
2
],
dtype
=
np
.
int32
)
})
refined_box_encodings
=
np
.
zeros
(
with
self
.
test_session
()
as
sess
:
[
total_num_padded_proposals
,
model
.
num_classes
,
4
])
detections_out
=
sess
.
run
(
detections
)
class_predictions_with_background
=
np
.
ones
(
[
total_num_padded_proposals
,
model
.
num_classes
+
1
])
num_proposals_placeholder
=
tf
.
placeholder
(
tf
.
int32
,
shape
=
num_proposals_shape
)
refined_box_encodings_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
refined_box_encoding_shape
)
class_predictions_with_background_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
class_predictions_with_background_shape
)
proposal_boxes_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
proposal_boxes_shape
)
image_shape_placeholder
=
tf
.
placeholder
(
tf
.
int32
,
shape
=
(
4
))
detections
=
model
.
postprocess
({
'refined_box_encodings'
:
refined_box_encodings_placeholder
,
'class_predictions_with_background'
:
class_predictions_with_background_placeholder
,
'num_proposals'
:
num_proposals_placeholder
,
'proposal_boxes'
:
proposal_boxes_placeholder
,
'image_shape'
:
image_shape_placeholder
,
})
with
self
.
test_session
(
graph
=
tf_graph
)
as
sess
:
detections_out
=
sess
.
run
(
detections
,
feed_dict
=
{
refined_box_encodings_placeholder
:
refined_box_encodings
,
class_predictions_with_background_placeholder
:
class_predictions_with_background
,
num_proposals_placeholder
:
num_proposals
,
proposal_boxes_placeholder
:
proposal_boxes
,
image_shape_placeholder
:
image_shape
})
self
.
assertAllEqual
(
detections_out
[
'detection_boxes'
].
shape
,
[
2
,
5
,
4
])
self
.
assertAllEqual
(
detections_out
[
'detection_boxes'
].
shape
,
[
2
,
5
,
4
])
self
.
assertAllClose
(
detections_out
[
'detection_scores'
],
self
.
assertAllClose
(
detections_out
[
'detection_scores'
],
[[
1
,
1
,
1
,
1
,
1
],
[
1
,
1
,
1
,
1
,
0
]])
[[
1
,
1
,
1
,
1
,
1
],
[
1
,
1
,
1
,
1
,
0
]])
...
@@ -571,6 +589,17 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
...
@@ -571,6 +589,17 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
[[
0
,
0
,
0
,
1
,
1
],
[
0
,
0
,
1
,
1
,
0
]])
[[
0
,
0
,
0
,
1
,
1
],
[
0
,
0
,
1
,
1
,
0
]])
self
.
assertAllClose
(
detections_out
[
'num_detections'
],
[
5
,
4
])
self
.
assertAllClose
(
detections_out
[
'num_detections'
],
[
5
,
4
])
def
test_preprocess_preserves_input_shapes
(
self
):
image_shapes
=
[(
3
,
None
,
None
,
3
),
(
None
,
10
,
10
,
3
),
(
None
,
None
,
None
,
3
)]
for
image_shape
in
image_shapes
:
model
=
self
.
_build_model
(
is_training
=
False
,
first_stage_only
=
False
,
second_stage_batch_size
=
6
)
image_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
image_shape
)
preprocessed_inputs
=
model
.
preprocess
(
image_placeholder
)
self
.
assertAllEqual
(
preprocessed_inputs
.
shape
.
as_list
(),
image_shape
)
def
test_loss_first_stage_only_mode
(
self
):
def
test_loss_first_stage_only_mode
(
self
):
model
=
self
.
_build_model
(
model
=
self
.
_build_model
(
is_training
=
True
,
first_stage_only
=
True
,
second_stage_batch_size
=
6
)
is_training
=
True
,
first_stage_only
=
True
,
second_stage_batch_size
=
6
)
...
@@ -957,7 +986,7 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
...
@@ -957,7 +986,7 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
exp_loc_loss
)
exp_loc_loss
)
self
.
assertAllClose
(
loss_dict_out
[
'second_stage_classification_loss'
],
0
)
self
.
assertAllClose
(
loss_dict_out
[
'second_stage_classification_loss'
],
0
)
def
test_restore_
fn
_classification
(
self
):
def
test_restore_
map_for
_classification
_ckpt
(
self
):
# Define mock tensorflow classification graph and save variables.
# Define mock tensorflow classification graph and save variables.
test_graph_classification
=
tf
.
Graph
()
test_graph_classification
=
tf
.
Graph
()
with
test_graph_classification
.
as_default
():
with
test_graph_classification
.
as_default
():
...
@@ -986,12 +1015,17 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
...
@@ -986,12 +1015,17 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
preprocessed_inputs
=
model
.
preprocess
(
inputs
)
preprocessed_inputs
=
model
.
preprocess
(
inputs
)
prediction_dict
=
model
.
predict
(
preprocessed_inputs
)
prediction_dict
=
model
.
predict
(
preprocessed_inputs
)
model
.
postprocess
(
prediction_dict
)
model
.
postprocess
(
prediction_dict
)
restore_fn
=
model
.
restore_fn
(
saved_model_path
,
var_map
=
model
.
restore_map
(
from_detection_checkpoint
=
False
)
from_detection_checkpoint
=
False
)
self
.
assertIsInstance
(
var_map
,
dict
)
saver
=
tf
.
train
.
Saver
(
var_map
)
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
restore_fn
(
sess
)
saver
.
restore
(
sess
,
saved_model_path
)
for
var
in
sess
.
run
(
tf
.
report_uninitialized_variables
()):
self
.
assertNotIn
(
model
.
first_stage_feature_extractor_scope
,
var
.
name
)
self
.
assertNotIn
(
model
.
second_stage_feature_extractor_scope
,
var
.
name
)
def
test_restore_
fn
_detection
(
self
):
def
test_restore_
map_for
_detection
_ckpt
(
self
):
# Define first detection graph and save variables.
# Define first detection graph and save variables.
test_graph_detection1
=
tf
.
Graph
()
test_graph_detection1
=
tf
.
Graph
()
with
test_graph_detection1
.
as_default
():
with
test_graph_detection1
.
as_default
():
...
@@ -1022,10 +1056,11 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
...
@@ -1022,10 +1056,11 @@ class FasterRCNNMetaArchTestBase(tf.test.TestCase):
preprocessed_inputs2
=
model2
.
preprocess
(
inputs2
)
preprocessed_inputs2
=
model2
.
preprocess
(
inputs2
)
prediction_dict2
=
model2
.
predict
(
preprocessed_inputs2
)
prediction_dict2
=
model2
.
predict
(
preprocessed_inputs2
)
model2
.
postprocess
(
prediction_dict2
)
model2
.
postprocess
(
prediction_dict2
)
restore_fn
=
model2
.
restore_fn
(
saved_model_path
,
var_map
=
model2
.
restore_map
(
from_detection_checkpoint
=
True
)
from_detection_checkpoint
=
True
)
self
.
assertIsInstance
(
var_map
,
dict
)
saver
=
tf
.
train
.
Saver
(
var_map
)
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
restore
_fn
(
sess
)
saver
.
restore
(
sess
,
saved_model_path
)
for
var
in
sess
.
run
(
tf
.
report_uninitialized_variables
()):
for
var
in
sess
.
run
(
tf
.
report_uninitialized_variables
()):
self
.
assertNotIn
(
model2
.
first_stage_feature_extractor_scope
,
var
.
name
)
self
.
assertNotIn
(
model2
.
first_stage_feature_extractor_scope
,
var
.
name
)
self
.
assertNotIn
(
model2
.
second_stage_feature_extractor_scope
,
self
.
assertNotIn
(
model2
.
second_stage_feature_extractor_scope
,
...
...
object_detection/meta_architectures/ssd_meta_arch.py
View file @
582bf927
...
@@ -23,13 +23,12 @@ from abc import abstractmethod
...
@@ -23,13 +23,12 @@ from abc import abstractmethod
import
re
import
re
import
tensorflow
as
tf
import
tensorflow
as
tf
from
object_detection.core
import
box_coder
as
bcoder
from
object_detection.core
import
box_list
from
object_detection.core
import
box_list
from
object_detection.core
import
box_predictor
as
bpredictor
from
object_detection.core
import
box_predictor
as
bpredictor
from
object_detection.core
import
model
from
object_detection.core
import
model
from
object_detection.core
import
standard_fields
as
fields
from
object_detection.core
import
standard_fields
as
fields
from
object_detection.core
import
target_assigner
from
object_detection.core
import
target_assigner
from
object_detection.utils
import
variables_helper
from
object_detection.utils
import
shape_utils
slim
=
tf
.
contrib
.
slim
slim
=
tf
.
contrib
.
slim
...
@@ -324,7 +323,8 @@ class SSDMetaArch(model.DetectionModel):
...
@@ -324,7 +323,8 @@ class SSDMetaArch(model.DetectionModel):
a list of pairs (height, width) for each feature map in feature_maps
a list of pairs (height, width) for each feature map in feature_maps
"""
"""
feature_map_shapes
=
[
feature_map_shapes
=
[
feature_map
.
get_shape
().
as_list
()
for
feature_map
in
feature_maps
shape_utils
.
combined_static_and_dynamic_shape
(
feature_map
)
for
feature_map
in
feature_maps
]
]
return
[(
shape
[
1
],
shape
[
2
])
for
shape
in
feature_map_shapes
]
return
[(
shape
[
1
],
shape
[
2
])
for
shape
in
feature_map_shapes
]
...
@@ -365,8 +365,7 @@ class SSDMetaArch(model.DetectionModel):
...
@@ -365,8 +365,7 @@ class SSDMetaArch(model.DetectionModel):
with
tf
.
name_scope
(
'Postprocessor'
):
with
tf
.
name_scope
(
'Postprocessor'
):
box_encodings
=
prediction_dict
[
'box_encodings'
]
box_encodings
=
prediction_dict
[
'box_encodings'
]
class_predictions
=
prediction_dict
[
'class_predictions_with_background'
]
class_predictions
=
prediction_dict
[
'class_predictions_with_background'
]
detection_boxes
=
bcoder
.
batch_decode
(
box_encodings
,
self
.
_box_coder
,
detection_boxes
=
self
.
_batch_decode
(
box_encodings
)
self
.
anchors
)
detection_boxes
=
tf
.
expand_dims
(
detection_boxes
,
axis
=
2
)
detection_boxes
=
tf
.
expand_dims
(
detection_boxes
,
axis
=
2
)
class_predictions_without_background
=
tf
.
slice
(
class_predictions
,
class_predictions_without_background
=
tf
.
slice
(
class_predictions
,
...
@@ -375,10 +374,14 @@ class SSDMetaArch(model.DetectionModel):
...
@@ -375,10 +374,14 @@ class SSDMetaArch(model.DetectionModel):
detection_scores
=
self
.
_score_conversion_fn
(
detection_scores
=
self
.
_score_conversion_fn
(
class_predictions_without_background
)
class_predictions_without_background
)
clip_window
=
tf
.
constant
([
0
,
0
,
1
,
1
],
tf
.
float32
)
clip_window
=
tf
.
constant
([
0
,
0
,
1
,
1
],
tf
.
float32
)
detections
=
self
.
_non_max_suppression_fn
(
detection_boxes
,
(
nmsed_boxes
,
nmsed_scores
,
nmsed_classes
,
_
,
detection_scores
,
num_detections
)
=
self
.
_non_max_suppression_fn
(
detection_boxes
,
clip_window
=
clip_window
)
detection_scores
,
return
detections
clip_window
=
clip_window
)
return
{
'detection_boxes'
:
nmsed_boxes
,
'detection_scores'
:
nmsed_scores
,
'detection_classes'
:
nmsed_classes
,
'num_detections'
:
tf
.
to_float
(
num_detections
)}
def
loss
(
self
,
prediction_dict
,
scope
=
None
):
def
loss
(
self
,
prediction_dict
,
scope
=
None
):
"""Compute scalar loss tensors with respect to provided groundtruth.
"""Compute scalar loss tensors with respect to provided groundtruth.
...
@@ -546,8 +549,7 @@ class SSDMetaArch(model.DetectionModel):
...
@@ -546,8 +549,7 @@ class SSDMetaArch(model.DetectionModel):
tf
.
slice
(
prediction_dict
[
'class_predictions_with_background'
],
tf
.
slice
(
prediction_dict
[
'class_predictions_with_background'
],
[
0
,
0
,
1
],
class_pred_shape
),
class_pred_shape
)
[
0
,
0
,
1
],
class_pred_shape
),
class_pred_shape
)
decoded_boxes
=
bcoder
.
batch_decode
(
prediction_dict
[
'box_encodings'
],
decoded_boxes
=
self
.
_batch_decode
(
prediction_dict
[
'box_encodings'
])
self
.
_box_coder
,
self
.
anchors
)
decoded_box_tensors_list
=
tf
.
unstack
(
decoded_boxes
)
decoded_box_tensors_list
=
tf
.
unstack
(
decoded_boxes
)
class_prediction_list
=
tf
.
unstack
(
class_predictions
)
class_prediction_list
=
tf
.
unstack
(
class_predictions
)
decoded_boxlist_list
=
[]
decoded_boxlist_list
=
[]
...
@@ -562,33 +564,51 @@ class SSDMetaArch(model.DetectionModel):
...
@@ -562,33 +564,51 @@ class SSDMetaArch(model.DetectionModel):
decoded_boxlist_list
=
decoded_boxlist_list
,
decoded_boxlist_list
=
decoded_boxlist_list
,
match_list
=
match_list
)
match_list
=
match_list
)
def
restore_fn
(
self
,
checkpoint_path
,
from_detection_checkpoint
=
True
):
def
_batch_decode
(
self
,
box_encodings
):
"""Return callable for loading a checkpoint into the tensorflow graph.
"""Decodes a batch of box encodings with respect to the anchors.
Args:
box_encodings: A float32 tensor of shape
[batch_size, num_anchors, box_code_size] containing box encodings.
Returns:
decoded_boxes: A float32 tensor of shape
[batch_size, num_anchors, 4] containing the decoded boxes.
"""
combined_shape
=
shape_utils
.
combined_static_and_dynamic_shape
(
box_encodings
)
batch_size
=
combined_shape
[
0
]
tiled_anchor_boxes
=
tf
.
tile
(
tf
.
expand_dims
(
self
.
anchors
.
get
(),
0
),
[
batch_size
,
1
,
1
])
tiled_anchors_boxlist
=
box_list
.
BoxList
(
tf
.
reshape
(
tiled_anchor_boxes
,
[
-
1
,
self
.
_box_coder
.
code_size
]))
decoded_boxes
=
self
.
_box_coder
.
decode
(
tf
.
reshape
(
box_encodings
,
[
-
1
,
self
.
_box_coder
.
code_size
]),
tiled_anchors_boxlist
)
return
tf
.
reshape
(
decoded_boxes
.
get
(),
tf
.
stack
([
combined_shape
[
0
],
combined_shape
[
1
],
4
]))
def
restore_map
(
self
,
from_detection_checkpoint
=
True
):
"""Returns a map of variables to load from a foreign checkpoint.
See parent class for details.
Args:
Args:
checkpoint_path: path to checkpoint to restore.
from_detection_checkpoint: whether to restore from a full detection
from_detection_checkpoint: whether to restore from a full detection
checkpoint (with compatible variable names) or to restore from a
checkpoint (with compatible variable names) or to restore from a
classification checkpoint for initialization prior to training.
classification checkpoint for initialization prior to training.
Returns:
Returns:
a callable which takes a tf.Session as input and loads a checkpoint whe
n
A dict mapping variable names (to load from a checkpoint) to variables i
n
run
.
the model graph
.
"""
"""
variables_to_restore
=
{}
variables_to_restore
=
{}
for
variable
in
tf
.
all_variables
():
for
variable
in
tf
.
all_variables
():
if
variable
.
op
.
name
.
startswith
(
self
.
_extract_features_scope
):
if
variable
.
op
.
name
.
startswith
(
self
.
_extract_features_scope
):
var_name
=
variable
.
op
.
name
var_name
=
variable
.
op
.
name
if
not
from_detection_checkpoint
:
if
not
from_detection_checkpoint
:
var_name
=
(
var_name
=
(
re
.
split
(
'^'
+
self
.
_extract_features_scope
+
'/'
,
re
.
split
(
'^'
+
self
.
_extract_features_scope
+
'/'
,
var_name
)[
-
1
])
var_name
)[
-
1
])
variables_to_restore
[
var_name
]
=
variable
variables_to_restore
[
var_name
]
=
variable
# TODO: Load variables selectively using scopes.
return
variables_to_restore
variables_to_restore
=
(
variables_helper
.
get_variables_available_in_checkpoint
(
variables_to_restore
,
checkpoint_path
))
saver
=
tf
.
train
.
Saver
(
variables_to_restore
)
def
restore
(
sess
):
saver
.
restore
(
sess
,
checkpoint_path
)
return
restore
object_detection/meta_architectures/ssd_meta_arch_test.py
View file @
582bf927
...
@@ -116,24 +116,46 @@ class SsdMetaArchTest(tf.test.TestCase):
...
@@ -116,24 +116,46 @@ class SsdMetaArchTest(tf.test.TestCase):
localization_loss_weight
,
normalize_loss_by_num_matches
,
localization_loss_weight
,
normalize_loss_by_num_matches
,
hard_example_miner
)
hard_example_miner
)
def
test_preprocess_preserves_input_shapes
(
self
):
image_shapes
=
[(
3
,
None
,
None
,
3
),
(
None
,
10
,
10
,
3
),
(
None
,
None
,
None
,
3
)]
for
image_shape
in
image_shapes
:
image_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
image_shape
)
preprocessed_inputs
=
self
.
_model
.
preprocess
(
image_placeholder
)
self
.
assertAllEqual
(
preprocessed_inputs
.
shape
.
as_list
(),
image_shape
)
def
test_predict_results_have_correct_keys_and_shapes
(
self
):
def
test_predict_results_have_correct_keys_and_shapes
(
self
):
batch_size
=
3
batch_size
=
3
preprocessed_input
=
tf
.
random_uniform
((
batch_size
,
2
,
2
,
3
),
image_size
=
2
dtype
=
tf
.
float32
)
input_shapes
=
[(
batch_size
,
image_size
,
image_size
,
3
),
prediction_dict
=
self
.
_model
.
predict
(
preprocessed_input
)
(
None
,
image_size
,
image_size
,
3
),
(
batch_size
,
None
,
None
,
3
),
self
.
assertTrue
(
'box_encodings'
in
prediction_dict
)
(
None
,
None
,
None
,
3
)]
self
.
assertTrue
(
'class_predictions_with_background'
in
prediction_dict
)
self
.
assertTrue
(
'feature_maps'
in
prediction_dict
)
expected_box_encodings_shape_out
=
(
expected_box_encodings_shape_out
=
(
batch_size
,
self
.
_num_anchors
,
self
.
_code_size
)
batch_size
,
self
.
_num_anchors
,
self
.
_code_size
)
expected_class_predictions_with_background_shape_out
=
(
expected_class_predictions_with_background_shape_out
=
(
batch_size
,
self
.
_num_anchors
,
self
.
_num_classes
+
1
)
batch_size
,
self
.
_num_anchors
,
self
.
_num_classes
+
1
)
init_op
=
tf
.
global_variables_initializer
()
with
self
.
test_session
()
as
sess
:
for
input_shape
in
input_shapes
:
sess
.
run
(
init_op
)
tf_graph
=
tf
.
Graph
()
prediction_out
=
sess
.
run
(
prediction_dict
)
with
tf_graph
.
as_default
():
preprocessed_input_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
input_shape
)
prediction_dict
=
self
.
_model
.
predict
(
preprocessed_input_placeholder
)
self
.
assertTrue
(
'box_encodings'
in
prediction_dict
)
self
.
assertTrue
(
'class_predictions_with_background'
in
prediction_dict
)
self
.
assertTrue
(
'feature_maps'
in
prediction_dict
)
init_op
=
tf
.
global_variables_initializer
()
with
self
.
test_session
(
graph
=
tf_graph
)
as
sess
:
sess
.
run
(
init_op
)
prediction_out
=
sess
.
run
(
prediction_dict
,
feed_dict
=
{
preprocessed_input_placeholder
:
np
.
random
.
uniform
(
size
=
(
batch_size
,
2
,
2
,
3
))})
self
.
assertAllEqual
(
prediction_out
[
'box_encodings'
].
shape
,
self
.
assertAllEqual
(
prediction_out
[
'box_encodings'
].
shape
,
expected_box_encodings_shape_out
)
expected_box_encodings_shape_out
)
self
.
assertAllEqual
(
self
.
assertAllEqual
(
...
@@ -142,10 +164,11 @@ class SsdMetaArchTest(tf.test.TestCase):
...
@@ -142,10 +164,11 @@ class SsdMetaArchTest(tf.test.TestCase):
def
test_postprocess_results_are_correct
(
self
):
def
test_postprocess_results_are_correct
(
self
):
batch_size
=
2
batch_size
=
2
preprocessed_input
=
tf
.
random_uniform
((
batch_size
,
2
,
2
,
3
),
image_size
=
2
dtype
=
tf
.
float32
)
input_shapes
=
[(
batch_size
,
image_size
,
image_size
,
3
),
prediction_dict
=
self
.
_model
.
predict
(
preprocessed_input
)
(
None
,
image_size
,
image_size
,
3
),
detections
=
self
.
_model
.
postprocess
(
prediction_dict
)
(
batch_size
,
None
,
None
,
3
),
(
None
,
None
,
None
,
3
)]
expected_boxes
=
np
.
array
([[[
0
,
0
,
.
5
,
.
5
],
expected_boxes
=
np
.
array
([[[
0
,
0
,
.
5
,
.
5
],
[
0
,
.
5
,
.
5
,
1
],
[
0
,
.
5
,
.
5
,
1
],
...
@@ -163,15 +186,25 @@ class SsdMetaArchTest(tf.test.TestCase):
...
@@ -163,15 +186,25 @@ class SsdMetaArchTest(tf.test.TestCase):
[
0
,
0
,
0
,
0
,
0
]])
[
0
,
0
,
0
,
0
,
0
]])
expected_num_detections
=
np
.
array
([
4
,
4
])
expected_num_detections
=
np
.
array
([
4
,
4
])
self
.
assertTrue
(
'detection_boxes'
in
detections
)
for
input_shape
in
input_shapes
:
self
.
assertTrue
(
'detection_scores'
in
detections
)
tf_graph
=
tf
.
Graph
()
self
.
assertTrue
(
'detection_classes'
in
detections
)
with
tf_graph
.
as_default
():
self
.
assertTrue
(
'num_detections'
in
detections
)
preprocessed_input_placeholder
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
input_shape
)
init_op
=
tf
.
global_variables_initializer
()
prediction_dict
=
self
.
_model
.
predict
(
preprocessed_input_placeholder
)
with
self
.
test_session
()
as
sess
:
detections
=
self
.
_model
.
postprocess
(
prediction_dict
)
sess
.
run
(
init_op
)
self
.
assertTrue
(
'detection_boxes'
in
detections
)
detections_out
=
sess
.
run
(
detections
)
self
.
assertTrue
(
'detection_scores'
in
detections
)
self
.
assertTrue
(
'detection_classes'
in
detections
)
self
.
assertTrue
(
'num_detections'
in
detections
)
init_op
=
tf
.
global_variables_initializer
()
with
self
.
test_session
(
graph
=
tf_graph
)
as
sess
:
sess
.
run
(
init_op
)
detections_out
=
sess
.
run
(
detections
,
feed_dict
=
{
preprocessed_input_placeholder
:
np
.
random
.
uniform
(
size
=
(
batch_size
,
2
,
2
,
3
))})
self
.
assertAllClose
(
detections_out
[
'detection_boxes'
],
expected_boxes
)
self
.
assertAllClose
(
detections_out
[
'detection_boxes'
],
expected_boxes
)
self
.
assertAllClose
(
detections_out
[
'detection_scores'
],
expected_scores
)
self
.
assertAllClose
(
detections_out
[
'detection_scores'
],
expected_scores
)
self
.
assertAllClose
(
detections_out
[
'detection_classes'
],
expected_classes
)
self
.
assertAllClose
(
detections_out
[
'detection_classes'
],
expected_classes
)
...
@@ -207,20 +240,21 @@ class SsdMetaArchTest(tf.test.TestCase):
...
@@ -207,20 +240,21 @@ class SsdMetaArchTest(tf.test.TestCase):
self
.
assertAllClose
(
losses_out
[
'classification_loss'
],
self
.
assertAllClose
(
losses_out
[
'classification_loss'
],
expected_classification_loss
)
expected_classification_loss
)
def
test_restore_
fn
_detection
(
self
):
def
test_restore_
map_for
_detection
_ckpt
(
self
):
init_op
=
tf
.
global_variables_initializer
()
init_op
=
tf
.
global_variables_initializer
()
saver
=
tf_saver
.
Saver
()
saver
=
tf_saver
.
Saver
()
save_path
=
self
.
get_temp_dir
()
save_path
=
self
.
get_temp_dir
()
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
sess
.
run
(
init_op
)
sess
.
run
(
init_op
)
saved_model_path
=
saver
.
save
(
sess
,
save_path
)
saved_model_path
=
saver
.
save
(
sess
,
save_path
)
restore_fn
=
self
.
_model
.
restore_fn
(
saved_model_path
,
var_map
=
self
.
_model
.
restore_map
(
from_detection_checkpoint
=
True
)
from_detection_checkpoint
=
True
)
self
.
assertIsInstance
(
var_map
,
dict
)
restore_fn
(
sess
)
saver
=
tf
.
train
.
Saver
(
var_map
)
saver
.
restore
(
sess
,
saved_model_path
)
for
var
in
sess
.
run
(
tf
.
report_uninitialized_variables
()):
for
var
in
sess
.
run
(
tf
.
report_uninitialized_variables
()):
self
.
assertNotIn
(
'FeatureExtractor'
,
var
.
name
)
self
.
assertNotIn
(
'FeatureExtractor'
,
var
.
name
)
def
test_restore_
fn
_classification
(
self
):
def
test_restore_
map_for
_classification
_ckpt
(
self
):
# Define mock tensorflow classification graph and save variables.
# Define mock tensorflow classification graph and save variables.
test_graph_classification
=
tf
.
Graph
()
test_graph_classification
=
tf
.
Graph
()
with
test_graph_classification
.
as_default
():
with
test_graph_classification
.
as_default
():
...
@@ -246,10 +280,11 @@ class SsdMetaArchTest(tf.test.TestCase):
...
@@ -246,10 +280,11 @@ class SsdMetaArchTest(tf.test.TestCase):
preprocessed_inputs
=
self
.
_model
.
preprocess
(
inputs
)
preprocessed_inputs
=
self
.
_model
.
preprocess
(
inputs
)
prediction_dict
=
self
.
_model
.
predict
(
preprocessed_inputs
)
prediction_dict
=
self
.
_model
.
predict
(
preprocessed_inputs
)
self
.
_model
.
postprocess
(
prediction_dict
)
self
.
_model
.
postprocess
(
prediction_dict
)
restore_fn
=
self
.
_model
.
restore_fn
(
saved_model_path
,
var_map
=
self
.
_model
.
restore_map
(
from_detection_checkpoint
=
False
)
from_detection_checkpoint
=
False
)
self
.
assertIsInstance
(
var_map
,
dict
)
saver
=
tf
.
train
.
Saver
(
var_map
)
with
self
.
test_session
()
as
sess
:
with
self
.
test_session
()
as
sess
:
restore
_fn
(
sess
)
saver
.
restore
(
sess
,
saved_model_path
)
for
var
in
sess
.
run
(
tf
.
report_uninitialized_variables
()):
for
var
in
sess
.
run
(
tf
.
report_uninitialized_variables
()):
self
.
assertNotIn
(
'FeatureExtractor'
,
var
.
name
)
self
.
assertNotIn
(
'FeatureExtractor'
,
var
.
name
)
...
...
object_detection/models/BUILD
View file @
582bf927
...
@@ -94,7 +94,6 @@ py_library(
...
@@ -94,7 +94,6 @@ py_library(
deps
=
[
deps
=
[
"//tensorflow"
,
"//tensorflow"
,
"//tensorflow_models/object_detection/meta_architectures:faster_rcnn_meta_arch"
,
"//tensorflow_models/object_detection/meta_architectures:faster_rcnn_meta_arch"
,
"//tensorflow_models/object_detection/utils:variables_helper"
,
"//tensorflow_models/slim:inception_resnet_v2"
,
"//tensorflow_models/slim:inception_resnet_v2"
,
],
],
)
)
...
...
object_detection/models/faster_rcnn_inception_resnet_v2_feature_extractor.py
View file @
582bf927
...
@@ -25,7 +25,6 @@ Huang et al. (https://arxiv.org/abs/1611.10012)
...
@@ -25,7 +25,6 @@ Huang et al. (https://arxiv.org/abs/1611.10012)
import
tensorflow
as
tf
import
tensorflow
as
tf
from
object_detection.meta_architectures
import
faster_rcnn_meta_arch
from
object_detection.meta_architectures
import
faster_rcnn_meta_arch
from
object_detection.utils
import
variables_helper
from
nets
import
inception_resnet_v2
from
nets
import
inception_resnet_v2
slim
=
tf
.
contrib
.
slim
slim
=
tf
.
contrib
.
slim
...
@@ -168,30 +167,30 @@ class FasterRCNNInceptionResnetV2FeatureExtractor(
...
@@ -168,30 +167,30 @@ class FasterRCNNInceptionResnetV2FeatureExtractor(
def
restore_from_classification_checkpoint_fn
(
def
restore_from_classification_checkpoint_fn
(
self
,
self
,
checkpoint_path
,
first_stage_feature_extractor_scope
,
first_stage_feature_extractor_scope
,
second_stage_feature_extractor_scope
):
second_stage_feature_extractor_scope
):
"""Returns
callable for loading a checkpoint into the tensorflow graph
.
"""Returns
a map of variables to load from a foreign checkpoint
.
Note that this overrides the default implementation in
Note that this overrides the default implementation in
faster_rcnn_meta_arch.FasterRCNNFeatureExtractor which does not work for
faster_rcnn_meta_arch.FasterRCNNFeatureExtractor which does not work for
InceptionResnetV2 checkpoints.
InceptionResnetV2 checkpoints.
TODO: revisit whether it's possible to force the `Repeat` namescope as
TODO: revisit whether it's possible to force the
created in `_extract_box_classifier_features` to start counting at 2 (e.g.
`Repeat` namescope as created in `_extract_box_classifier_features` to
`Repeat_2`) so that the default restore_fn can be used.
start counting at 2 (e.g. `Repeat_2`) so that the default restore_fn can
be used.
Args:
Args:
checkpoint_path: Path to checkpoint to restore.
first_stage_feature_extractor_scope: A scope name for the first stage
first_stage_feature_extractor_scope: A scope name for the first stage
feature extractor.
feature extractor.
second_stage_feature_extractor_scope: A scope name for the second stage
second_stage_feature_extractor_scope: A scope name for the second stage
feature extractor.
feature extractor.
Returns:
Returns:
a callable which takes a tf.Session as input and loads a checkpoint whe
n
A dict mapping variable names (to load from a checkpoint) to variables i
n
run
.
the model graph
.
"""
"""
variables_to_restore
=
{}
variables_to_restore
=
{}
for
variable
in
tf
.
global_variables
():
for
variable
in
tf
.
global_variables
():
if
variable
.
op
.
name
.
startswith
(
if
variable
.
op
.
name
.
startswith
(
...
@@ -207,10 +206,4 @@ class FasterRCNNInceptionResnetV2FeatureExtractor(
...
@@ -207,10 +206,4 @@ class FasterRCNNInceptionResnetV2FeatureExtractor(
var_name
=
var_name
.
replace
(
var_name
=
var_name
.
replace
(
second_stage_feature_extractor_scope
+
'/'
,
''
)
second_stage_feature_extractor_scope
+
'/'
,
''
)
variables_to_restore
[
var_name
]
=
variable
variables_to_restore
[
var_name
]
=
variable
variables_to_restore
=
(
return
variables_to_restore
variables_helper
.
get_variables_available_in_checkpoint
(
variables_to_restore
,
checkpoint_path
))
saver
=
tf
.
train
.
Saver
(
variables_to_restore
)
def
restore
(
sess
):
saver
.
restore
(
sess
,
checkpoint_path
)
return
restore
object_detection/trainer.py
View file @
582bf927
...
@@ -211,9 +211,15 @@ def train(create_tensor_dict_fn, create_model_fn, train_config, master, task,
...
@@ -211,9 +211,15 @@ def train(create_tensor_dict_fn, create_model_fn, train_config, master, task,
# Create ops required to initialize the model from a given checkpoint.
# Create ops required to initialize the model from a given checkpoint.
init_fn
=
None
init_fn
=
None
if
train_config
.
fine_tune_checkpoint
:
if
train_config
.
fine_tune_checkpoint
:
init_fn
=
detection_model
.
restore_fn
(
var_map
=
detection_model
.
restore_map
(
train_config
.
fine_tune_checkpoint
,
from_detection_checkpoint
=
train_config
.
from_detection_checkpoint
)
from_detection_checkpoint
=
train_config
.
from_detection_checkpoint
)
available_var_map
=
(
variables_helper
.
get_variables_available_in_checkpoint
(
var_map
,
train_config
.
fine_tune_checkpoint
))
init_saver
=
tf
.
train
.
Saver
(
available_var_map
)
def
initializer_fn
(
sess
):
init_saver
.
restore
(
sess
,
train_config
.
fine_tune_checkpoint
)
init_fn
=
initializer_fn
with
tf
.
device
(
deploy_config
.
optimizer_device
()):
with
tf
.
device
(
deploy_config
.
optimizer_device
()):
total_loss
,
grads_and_vars
=
model_deploy
.
optimize_clones
(
total_loss
,
grads_and_vars
=
model_deploy
.
optimize_clones
(
...
...
object_detection/trainer_test.py
View file @
582bf927
...
@@ -139,21 +139,18 @@ class FakeDetectionModel(model.DetectionModel):
...
@@ -139,21 +139,18 @@ class FakeDetectionModel(model.DetectionModel):
}
}
return
loss_dict
return
loss_dict
def
restore_
fn
(
self
,
checkpoint_path
,
from_detection_checkpoint
=
True
):
def
restore_
map
(
self
,
from_detection_checkpoint
=
True
):
"""Return
callable for loading a checkpoint into the tensorflow graph
.
"""Return
s a map of variables to load from a foreign checkpoint
.
Args:
Args:
checkpoint_path: path to checkpoint to restore.
from_detection_checkpoint: whether to restore from a full detection
from_detection_checkpoint: whether to restore from a full detection
checkpoint (with compatible variable names) or to restore from a
checkpoint (with compatible variable names) or to restore from a
classification checkpoint for initialization prior to training.
classification checkpoint for initialization prior to training.
Returns:
Returns:
a callable which takes a tf.Session and does nothing
.
A dict mapping variable names to variables
.
"""
"""
def
restore
(
unused_sess
):
return
{
var
.
op
.
name
:
var
for
var
in
tf
.
global_variables
()}
return
return
restore
class
TrainerTest
(
tf
.
test
.
TestCase
):
class
TrainerTest
(
tf
.
test
.
TestCase
):
...
...
object_detection/utils/BUILD
View file @
582bf927
...
@@ -120,6 +120,7 @@ py_library(
...
@@ -120,6 +120,7 @@ py_library(
"//tensorflow_models/object_detection/core:box_list"
,
"//tensorflow_models/object_detection/core:box_list"
,
"//tensorflow_models/object_detection/core:box_predictor"
,
"//tensorflow_models/object_detection/core:box_predictor"
,
"//tensorflow_models/object_detection/core:matcher"
,
"//tensorflow_models/object_detection/core:matcher"
,
"//tensorflow_models/object_detection/utils:shape_utils"
],
],
)
)
...
...
Prev
1
2
Next
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment