Skip to content

GitLab

Commit 9fce9c64 authored by Zhichao Lu's avatar Zhichao Lu Committed by pkulzc
Browse files

Merged commit includes the following changes: 

199348852  by Zhichao Lu:

    Small typos fixes in VRD evaluation.

--
199315191  by Zhichao Lu:

    Change padding shapes when additional channels are available.

--
199309180  by Zhichao Lu:

    Adds minor fixes to the Object Detection API implementation.

--
199298605  by Zhichao Lu:

    Force num_readers to be 1 when only input file is not sharded.

--
199292952  by Zhichao Lu:

    Adds image-level labels parsing into TfExampleDetectionAndGTParser.

--
199259866  by Zhichao Lu:

    Visual Relationships Evaluation executable.

--
199208330  by Zhichao Lu:

    Infer train_config.batch_size as the effective batch size. Therefore we need to divide the effective batch size in trainer by train_config.replica_to_aggregate to get per worker batch size.

--
199207842  by Zhichao Lu:

    Internal change.

--
199204222  by Zhichao Lu:

    In case the image has more than three channels, we only take the first three channels for visualization.

--
199194388  by Zhichao Lu:

    Correcting protocols description: VOC 2007 -> VOC 2012.

--
199188290  by Zhichao Lu:

    Adds per-relationship APs and mAP computation to VRD evaluation.

--
199158801  by Zhichao Lu:

    If available, additional channels are merged with input image.

--
199099637  by Zhichao Lu:

    OpenImages Challenge metric support:
    -adding verified labels standard field for TFExample;
    -adding tfrecord creation functionality.

--
198957391  by Zhichao Lu:

    Allow tf record sharding when creating pets dataset.

--
198925184  by Zhichao Lu:

    Introduce moving average support for evaluation. Also adding the ability to override this configuration via config_util.

--
198918186  by Zhichao Lu:

    Handles the case where there are 0 box masks.

--
198809009  by Zhichao Lu:

    Plumb groundtruth weights into target assigner for Faster RCNN.

--
198759987  by Zhichao Lu:

    Fix object detection test broken by shape inference.

--
198668602  by Zhichao Lu:

    Adding a new input field in data_decoders/tf_example_decoder.py for storing additional channels.

--
198530013  by Zhichao Lu:

    An util for hierarchical expandion of boxes and labels of OID dataset.

--
198503124  by Zhichao Lu:

    Fix dimension mismatch error introduced by
    https://github.com/tensorflow/tensorflow/pull/18251, or cl/194031845.
    After above change, conv2d strictly checks for conv_dims + 2 == input_rank.

--
198445807  by Zhichao Lu:

    Enabling Object Detection Challenge 2018 metric in evaluator.py framework for
    running eval job.
    Renaming old OpenImages V2 metric.

--
198413950  by Zhichao Lu:

    Support generic configuration override using namespaced keys

    Useful for adding custom hyper-parameter tuning fields without having to add custom override methods to config_utils.py.

--
198106437  by Zhichao Lu:

    Enable fused batchnorm now that quantization is supported.

--
198048364  by Zhichao Lu:

    Add support for keypoints in tf sequence examples and some util ops.

--
198004736  by Zhichao Lu:

    Relax postprocessing unit tests that are based on assumption that tf.image.non_max_suppression are stable with respect to input.

--
197997513  by Zhichao Lu:

    More lenient validation for normalized box boundaries.

--
197940068  by Zhichao Lu:

    A couple of minor updates/fixes:
    - Updating input reader proto with option to use display_name when decoding data.
    - Updating visualization tool to specify whether using absolute or normalized box coordinates. Appropriate boxes will now appear in TB when using model_main.py

--
197920152  by Zhichao Lu:

    Add quantized training support in the new OD binaries and a config for SSD Mobilenet v1 quantized training that is TPU compatible.

--
197213563  by Zhichao Lu:

    Do not share batch_norm for classification and regression tower in weight shared box predictor.

--
197196757  by Zhichao Lu:

    Relax the box_predictor api to return box_prediction of shape [batch_size, num_anchors, code_size] in addition to [batch_size, num_anchors, (1|q), code_size].

--
196898361  by Zhichao Lu:

    Allow per-channel scalar value to pad input image with when using keep aspect ratio resizer (when pad_to_max_dimension=True).

    In Object Detection Pipeline, we pad image before normalization and this skews batch_norm statistics during training. The option to set per channel pad value lets us truly pad with zeros.

--
196592101  by Zhichao Lu:

    Fix bug regarding tfrecord shuffling in object_detection

--
196320138  by Zhichao Lu:

    Fix typo in exporting_models.md

--

PiperOrigin-RevId: 199348852
parent ed901b73
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 3240 additions and 1566 deletions
research/object_detection/builders/dataset_builder.py
View file @ 9fce9c64
......@@ -56,15 +56,26 @@ def _get_padding_shapes(dataset, max_num_boxes=None, num_classes=None,
else:
height, width = spatial_image_shape # pylint: disable=unpacking-non-sequence
num_additional_channels = 0
if fields.InputDataFields.image_additional_channels in dataset.output_shapes:
num_additional_channels = dataset.output_shapes[
fields.InputDataFields.image_additional_channels].dims[2].value
padding_shapes = {
fields.InputDataFields.image: [height, width, 3],
# Additional channels are merged before batching.
fields.InputDataFields.image: [
height, width, 3 + num_additional_channels
],
fields.InputDataFields.image_additional_channels: [
height, width, num_additional_channels
],
fields.InputDataFields.source_id: [],
fields.InputDataFields.filename: [],
fields.InputDataFields.key: [],
fields.InputDataFields.groundtruth_difficult: [max_num_boxes],
fields.InputDataFields.groundtruth_boxes: [max_num_boxes, 4],
fields.InputDataFields.groundtruth_instance_masks: [max_num_boxes, height,
width],
fields.InputDataFields.groundtruth_instance_masks: [
max_num_boxes, height, width
],
fields.InputDataFields.groundtruth_is_crowd: [max_num_boxes],
fields.InputDataFields.groundtruth_group_of: [max_num_boxes],
fields.InputDataFields.groundtruth_area: [max_num_boxes],
......@@ -74,7 +85,8 @@ def _get_padding_shapes(dataset, max_num_boxes=None, num_classes=None,
fields.InputDataFields.groundtruth_label_scores: [max_num_boxes],
fields.InputDataFields.true_image_shape: [3],
fields.InputDataFields.multiclass_scores: [
max_num_boxes, num_classes + 1 if num_classes is not None else None],
max_num_boxes, num_classes + 1 if num_classes is not None else None
],
}
# Determine whether groundtruth_classes are integers or one-hot encodings, and
# apply batching appropriately.
......@@ -90,7 +102,9 @@ def _get_padding_shapes(dataset, max_num_boxes=None, num_classes=None,
'rank 2 tensor (one-hot encodings)')
if fields.InputDataFields.original_image in dataset.output_shapes:
padding_shapes[fields.InputDataFields.original_image] = [None, None, 3]
padding_shapes[fields.InputDataFields.original_image] = [
None, None, 3 + num_additional_channels
]
if fields.InputDataFields.groundtruth_keypoints in dataset.output_shapes:
tensor_shape = dataset.output_shapes[fields.InputDataFields.
groundtruth_keypoints]
......@@ -108,9 +122,13 @@ def _get_padding_shapes(dataset, max_num_boxes=None, num_classes=None,
for tensor_key, _ in dataset.output_shapes.items()}
def build(input_reader_config, transform_input_data_fn=None,
batch_size=None, max_num_boxes=None, num_classes=None,
spatial_image_shape=None):
def build(input_reader_config,
transform_input_data_fn=None,
batch_size=None,
max_num_boxes=None,
num_classes=None,
spatial_image_shape=None,
num_additional_channels=0):
"""Builds a tf.data.Dataset.
Builds a tf.data.Dataset by applying the `transform_input_data_fn` on all
......@@ -128,6 +146,7 @@ def build(input_reader_config, transform_input_data_fn=None,
spatial_image_shape: A list of two integers of the form [height, width]
containing expected spatial shape of the image after applying
transform_input_data_fn. If None, will use dynamic shapes.
num_additional_channels: Number of additional channels to use in the input.
Returns:
A tf.data.Dataset based on the input_reader_config.
......@@ -152,7 +171,9 @@ def build(input_reader_config, transform_input_data_fn=None,
decoder = tf_example_decoder.TfExampleDecoder(
load_instance_masks=input_reader_config.load_instance_masks,
instance_mask_type=input_reader_config.mask_type,
label_map_proto_file=label_map_proto_file)
label_map_proto_file=label_map_proto_file,
use_display_name=input_reader_config.use_display_name,
num_additional_channels=num_additional_channels)
def process_fn(value):
processed = decoder.decode(value)
......
research/object_detection/builders/dataset_builder_test.py
View file @ 9fce9c64
......@@ -30,49 +30,50 @@ from object_detection.utils import dataset_util
class DatasetBuilderTest(tf.test.TestCase):
def create_tf_record(self):
def create_tf_record(self, has_additional_channels=False):
path = os.path.join(self.get_temp_dir(), 'tfrecord')
writer = tf.python_io.TFRecordWriter(path)
image_tensor = np.random.randint(255, size=(4, 5, 3)).astype(np.uint8)
additional_channels_tensor = np.random.randint(
255, size=(4, 5, 1)).astype(np.uint8)
flat_mask = (4 * 5) * [1.0]
with self.test_session():
encoded_jpeg = tf.image.encode_jpeg(tf.constant(image_tensor)).eval()
encoded_additional_channels_jpeg = tf.image.encode_jpeg(
tf.constant(additional_channels_tensor)).eval()
features = {
'image/encoded':
feature_pb2.Feature(
bytes_list=feature_pb2.BytesList(value=[encoded_jpeg])),
'image/format':
feature_pb2.Feature(
bytes_list=feature_pb2.BytesList(value=['jpeg'.encode('utf-8')])
),
'image/height':
feature_pb2.Feature(int64_list=feature_pb2.Int64List(value=[4])),
'image/width':
feature_pb2.Feature(int64_list=feature_pb2.Int64List(value=[5])),
'image/object/bbox/xmin':
feature_pb2.Feature(float_list=feature_pb2.FloatList(value=[0.0])),
'image/object/bbox/xmax':
feature_pb2.Feature(float_list=feature_pb2.FloatList(value=[1.0])),
'image/object/bbox/ymin':
feature_pb2.Feature(float_list=feature_pb2.FloatList(value=[0.0])),
'image/object/bbox/ymax':
feature_pb2.Feature(float_list=feature_pb2.FloatList(value=[1.0])),
'image/object/class/label':
feature_pb2.Feature(int64_list=feature_pb2.Int64List(value=[2])),
'image/object/mask':
feature_pb2.Feature(
float_list=feature_pb2.FloatList(value=flat_mask)),
}
if has_additional_channels:
features['image/additional_channels/encoded'] = feature_pb2.Feature(
bytes_list=feature_pb2.BytesList(
value=[encoded_additional_channels_jpeg] * 2))
example = example_pb2.Example(
features=feature_pb2.Features(
feature={
'image/encoded':
feature_pb2.Feature(
bytes_list=feature_pb2.BytesList(value=[encoded_jpeg])),
'image/format':
feature_pb2.Feature(
bytes_list=feature_pb2.BytesList(
value=['jpeg'.encode('utf-8')])),
'image/height':
feature_pb2.Feature(
int64_list=feature_pb2.Int64List(value=[4])),
'image/width':
feature_pb2.Feature(
int64_list=feature_pb2.Int64List(value=[5])),
'image/object/bbox/xmin':
feature_pb2.Feature(
float_list=feature_pb2.FloatList(value=[0.0])),
'image/object/bbox/xmax':
feature_pb2.Feature(
float_list=feature_pb2.FloatList(value=[1.0])),
'image/object/bbox/ymin':
feature_pb2.Feature(
float_list=feature_pb2.FloatList(value=[0.0])),
'image/object/bbox/ymax':
feature_pb2.Feature(
float_list=feature_pb2.FloatList(value=[1.0])),
'image/object/class/label':
feature_pb2.Feature(
int64_list=feature_pb2.Int64List(value=[2])),
'image/object/mask':
feature_pb2.Feature(
float_list=feature_pb2.FloatList(value=flat_mask)),
}))
features=feature_pb2.Features(feature=features))
writer.write(example.SerializeToString())
writer.close()
......@@ -218,6 +219,31 @@ class DatasetBuilderTest(tf.test.TestCase):
[2, 2, 4, 5],
output_dict[fields.InputDataFields.groundtruth_instance_masks].shape)
def test_build_tf_record_input_reader_with_additional_channels(self):
tf_record_path = self.create_tf_record(has_additional_channels=True)
input_reader_text_proto = """
shuffle: false
num_readers: 1
tf_record_input_reader {{
input_path: '{0}'
}}
""".format(tf_record_path)
input_reader_proto = input_reader_pb2.InputReader()
text_format.Merge(input_reader_text_proto, input_reader_proto)
tensor_dict = dataset_util.make_initializable_iterator(
dataset_builder.build(
input_reader_proto, batch_size=2,
num_additional_channels=2)).get_next()
sv = tf.train.Supervisor(logdir=self.get_temp_dir())
with sv.prepare_or_wait_for_session() as sess:
sv.start_queue_runners(sess)
output_dict = sess.run(tensor_dict)
self.assertEquals((2, 4, 5, 5),
output_dict[fields.InputDataFields.image].shape)
def test_raises_error_with_no_input_paths(self):
input_reader_text_proto = """
shuffle: false
......
research/object_detection/builders/image_resizer_builder.py
View file @ 9fce9c64
......@@ -79,12 +79,17 @@ def build(image_resizer_config):
keep_aspect_ratio_config.max_dimension):
raise ValueError('min_dimension > max_dimension')
method = _tf_resize_method(keep_aspect_ratio_config.resize_method)
per_channel_pad_value = (0, 0, 0)
if keep_aspect_ratio_config.per_channel_pad_value:
per_channel_pad_value = tuple(keep_aspect_ratio_config.
per_channel_pad_value)
image_resizer_fn = functools.partial(
preprocessor.resize_to_range,
min_dimension=keep_aspect_ratio_config.min_dimension,
max_dimension=keep_aspect_ratio_config.max_dimension,
method=method,
pad_to_max_dimension=keep_aspect_ratio_config.pad_to_max_dimension)
pad_to_max_dimension=keep_aspect_ratio_config.pad_to_max_dimension,
per_channel_pad_value=per_channel_pad_value)
if not keep_aspect_ratio_config.convert_to_grayscale:
return image_resizer_fn
elif image_resizer_oneof == 'fixed_shape_resizer':
......
research/object_detection/builders/image_resizer_builder_test.py
View file @ 9fce9c64
......@@ -52,6 +52,9 @@ class ImageResizerBuilderTest(tf.test.TestCase):
min_dimension: 10
max_dimension: 20
pad_to_max_dimension: true
per_channel_pad_value: 3
per_channel_pad_value: 4
per_channel_pad_value: 5
}
"""
input_shape = (50, 25, 3)
......
research/object_detection/builders/model_builder.py deleted 100644 → 0
View file @ ed901b73
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""A function to build a DetectionModel from configuration."""
from object_detection.builders import anchor_generator_builder
from object_detection.builders import box_coder_builder
from object_detection.builders import box_predictor_builder
from object_detection.builders import hyperparams_builder
from object_detection.builders import image_resizer_builder
from object_detection.builders import losses_builder
from object_detection.builders import matcher_builder
from object_detection.builders import post_processing_builder
from object_detection.builders import region_similarity_calculator_builder as sim_calc
from object_detection.core import box_predictor
from object_detection.meta_architectures import faster_rcnn_meta_arch
from object_detection.meta_architectures import rfcn_meta_arch
from object_detection.meta_architectures import ssd_meta_arch
from object_detection.models import faster_rcnn_inception_resnet_v2_feature_extractor as frcnn_inc_res
from object_detection.models import faster_rcnn_inception_v2_feature_extractor as frcnn_inc_v2
from object_detection.models import faster_rcnn_nas_feature_extractor as frcnn_nas
from object_detection.models import faster_rcnn_pnas_feature_extractor as frcnn_pnas
from object_detection.models import faster_rcnn_resnet_v1_feature_extractor as frcnn_resnet_v1
from object_detection.models import ssd_resnet_v1_fpn_feature_extractor as ssd_resnet_v1_fpn
from object_detection.models.embedded_ssd_mobilenet_v1_feature_extractor import EmbeddedSSDMobileNetV1FeatureExtractor
from object_detection.models.ssd_inception_v2_feature_extractor import SSDInceptionV2FeatureExtractor
from object_detection.models.ssd_inception_v3_feature_extractor import SSDInceptionV3FeatureExtractor
from object_detection.models.ssd_mobilenet_v1_feature_extractor import SSDMobileNetV1FeatureExtractor
from object_detection.models.ssd_mobilenet_v2_feature_extractor import SSDMobileNetV2FeatureExtractor
from object_detection.protos import model_pb2
# A map of names to SSD feature extractors.
SSD_FEATURE_EXTRACTOR_CLASS_MAP = {
'ssd_inception_v2': SSDInceptionV2FeatureExtractor,
'ssd_inception_v3': SSDInceptionV3FeatureExtractor,
'ssd_mobilenet_v1': SSDMobileNetV1FeatureExtractor,
'ssd_mobilenet_v2': SSDMobileNetV2FeatureExtractor,
'ssd_resnet50_v1_fpn': ssd_resnet_v1_fpn.SSDResnet50V1FpnFeatureExtractor,
'ssd_resnet101_v1_fpn': ssd_resnet_v1_fpn.SSDResnet101V1FpnFeatureExtractor,
'ssd_resnet152_v1_fpn': ssd_resnet_v1_fpn.SSDResnet152V1FpnFeatureExtractor,
'embedded_ssd_mobilenet_v1': EmbeddedSSDMobileNetV1FeatureExtractor,
}
# A map of names to Faster R-CNN feature extractors.
FASTER_RCNN_FEATURE_EXTRACTOR_CLASS_MAP = {
'faster_rcnn_nas':
frcnn_nas.FasterRCNNNASFeatureExtractor,
'faster_rcnn_pnas':
frcnn_pnas.FasterRCNNPNASFeatureExtractor,
'faster_rcnn_inception_resnet_v2':
frcnn_inc_res.FasterRCNNInceptionResnetV2FeatureExtractor,
'faster_rcnn_inception_v2':
frcnn_inc_v2.FasterRCNNInceptionV2FeatureExtractor,
'faster_rcnn_resnet50':
frcnn_resnet_v1.FasterRCNNResnet50FeatureExtractor,
'faster_rcnn_resnet101':
frcnn_resnet_v1.FasterRCNNResnet101FeatureExtractor,
'faster_rcnn_resnet152':
frcnn_resnet_v1.FasterRCNNResnet152FeatureExtractor,
}
def build(model_config, is_training, add_summaries=True,
add_background_class=True):
"""Builds a DetectionModel based on the model config.
Args:
model_config: A model.proto object containing the config for the desired
DetectionModel.
is_training: True if this model is being built for training purposes.
add_summaries: Whether to add tensorflow summaries in the model graph.
add_background_class: Whether to add an implicit background class to one-hot
encodings of groundtruth labels. Set to false if using groundtruth labels
with an explicit background class or using multiclass scores instead of
truth in the case of distillation. Ignored in the case of faster_rcnn.
Returns:
DetectionModel based on the config.
Raises:
ValueError: On invalid meta architecture or model.
"""
if not isinstance(model_config, model_pb2.DetectionModel):
raise ValueError('model_config not of type model_pb2.DetectionModel.')
meta_architecture = model_config.WhichOneof('model')
if meta_architecture == 'ssd':
return _build_ssd_model(model_config.ssd, is_training, add_summaries,
add_background_class)
if meta_architecture == 'faster_rcnn':
return _build_faster_rcnn_model(model_config.faster_rcnn, is_training,
add_summaries)
raise ValueError('Unknown meta architecture: {}'.format(meta_architecture))
def _build_ssd_feature_extractor(feature_extractor_config, is_training,
reuse_weights=None):
"""Builds a ssd_meta_arch.SSDFeatureExtractor based on config.
Args:
feature_extractor_config: A SSDFeatureExtractor proto config from ssd.proto.
is_training: True if this feature extractor is being built for training.
reuse_weights: if the feature extractor should reuse weights.
Returns:
ssd_meta_arch.SSDFeatureExtractor based on config.
Raises:
ValueError: On invalid feature extractor type.
"""
feature_type = feature_extractor_config.type
depth_multiplier = feature_extractor_config.depth_multiplier
min_depth = feature_extractor_config.min_depth
pad_to_multiple = feature_extractor_config.pad_to_multiple
use_explicit_padding = feature_extractor_config.use_explicit_padding
use_depthwise = feature_extractor_config.use_depthwise
conv_hyperparams = hyperparams_builder.build(
feature_extractor_config.conv_hyperparams, is_training)
override_base_feature_extractor_hyperparams = (
feature_extractor_config.override_base_feature_extractor_hyperparams)
if feature_type not in SSD_FEATURE_EXTRACTOR_CLASS_MAP:
raise ValueError('Unknown ssd feature_extractor: {}'.format(feature_type))
feature_extractor_class = SSD_FEATURE_EXTRACTOR_CLASS_MAP[feature_type]
return feature_extractor_class(
is_training, depth_multiplier, min_depth, pad_to_multiple,
conv_hyperparams, reuse_weights, use_explicit_padding, use_depthwise,
override_base_feature_extractor_hyperparams)
def _build_ssd_model(ssd_config, is_training, add_summaries,
add_background_class=True):
"""Builds an SSD detection model based on the model config.
Args:
ssd_config: A ssd.proto object containing the config for the desired
SSDMetaArch.
is_training: True if this model is being built for training purposes.
add_summaries: Whether to add tf summaries in the model.
add_background_class: Whether to add an implicit background class to one-hot
encodings of groundtruth labels. Set to false if using groundtruth labels
with an explicit background class or using multiclass scores instead of
truth in the case of distillation.
Returns:
SSDMetaArch based on the config.
Raises:
ValueError: If ssd_config.type is not recognized (i.e. not registered in
model_class_map).
"""
num_classes = ssd_config.num_classes
# Feature extractor
feature_extractor = _build_ssd_feature_extractor(
feature_extractor_config=ssd_config.feature_extractor,
is_training=is_training)
box_coder = box_coder_builder.build(ssd_config.box_coder)
matcher = matcher_builder.build(ssd_config.matcher)
region_similarity_calculator = sim_calc.build(
ssd_config.similarity_calculator)
encode_background_as_zeros = ssd_config.encode_background_as_zeros
negative_class_weight = ssd_config.negative_class_weight
ssd_box_predictor = box_predictor_builder.build(hyperparams_builder.build,
ssd_config.box_predictor,
is_training, num_classes)
anchor_generator = anchor_generator_builder.build(
ssd_config.anchor_generator)
image_resizer_fn = image_resizer_builder.build(ssd_config.image_resizer)
non_max_suppression_fn, score_conversion_fn = post_processing_builder.build(
ssd_config.post_processing)
(classification_loss, localization_loss, classification_weight,
localization_weight, hard_example_miner,
random_example_sampler) = losses_builder.build(ssd_config.loss)
normalize_loss_by_num_matches = ssd_config.normalize_loss_by_num_matches
normalize_loc_loss_by_codesize = ssd_config.normalize_loc_loss_by_codesize
return ssd_meta_arch.SSDMetaArch(
is_training,
anchor_generator,
ssd_box_predictor,
box_coder,
feature_extractor,
matcher,
region_similarity_calculator,
encode_background_as_zeros,
negative_class_weight,
image_resizer_fn,
non_max_suppression_fn,
score_conversion_fn,
classification_loss,
localization_loss,
classification_weight,
localization_weight,
normalize_loss_by_num_matches,
hard_example_miner,
add_summaries=add_summaries,
normalize_loc_loss_by_codesize=normalize_loc_loss_by_codesize,
freeze_batchnorm=ssd_config.freeze_batchnorm,
inplace_batchnorm_update=ssd_config.inplace_batchnorm_update,
add_background_class=add_background_class,
random_example_sampler=random_example_sampler)
def _build_faster_rcnn_feature_extractor(
feature_extractor_config, is_training, reuse_weights=None,
inplace_batchnorm_update=False):
"""Builds a faster_rcnn_meta_arch.FasterRCNNFeatureExtractor based on config.
Args:
feature_extractor_config: A FasterRcnnFeatureExtractor proto config from
faster_rcnn.proto.
is_training: True if this feature extractor is being built for training.
reuse_weights: if the feature extractor should reuse weights.
inplace_batchnorm_update: Whether to update batch_norm inplace during
training. This is required for batch norm to work correctly on TPUs. When
this is false, user must add a control dependency on
tf.GraphKeys.UPDATE_OPS for train/loss op in order to update the batch
norm moving average parameters.
Returns:
faster_rcnn_meta_arch.FasterRCNNFeatureExtractor based on config.
Raises:
ValueError: On invalid feature extractor type.
"""
if inplace_batchnorm_update:
raise ValueError('inplace batchnorm updates not supported.')
feature_type = feature_extractor_config.type
first_stage_features_stride = (
feature_extractor_config.first_stage_features_stride)
batch_norm_trainable = feature_extractor_config.batch_norm_trainable
if feature_type not in FASTER_RCNN_FEATURE_EXTRACTOR_CLASS_MAP:
raise ValueError('Unknown Faster R-CNN feature_extractor: {}'.format(
feature_type))
feature_extractor_class = FASTER_RCNN_FEATURE_EXTRACTOR_CLASS_MAP[
feature_type]
return feature_extractor_class(
is_training, first_stage_features_stride,
batch_norm_trainable, reuse_weights)
def _build_faster_rcnn_model(frcnn_config, is_training, add_summaries):
"""Builds a Faster R-CNN or R-FCN detection model based on the model config.
Builds R-FCN model if the second_stage_box_predictor in the config is of type
`rfcn_box_predictor` else builds a Faster R-CNN model.
Args:
frcnn_config: A faster_rcnn.proto object containing the config for the
desired FasterRCNNMetaArch or RFCNMetaArch.
is_training: True if this model is being built for training purposes.
add_summaries: Whether to add tf summaries in the model.
Returns:
FasterRCNNMetaArch based on the config.
Raises:
ValueError: If frcnn_config.type is not recognized (i.e. not registered in
model_class_map).
"""
num_classes = frcnn_config.num_classes
image_resizer_fn = image_resizer_builder.build(frcnn_config.image_resizer)
feature_extractor = _build_faster_rcnn_feature_extractor(
frcnn_config.feature_extractor, is_training,
frcnn_config.inplace_batchnorm_update)
number_of_stages = frcnn_config.number_of_stages
first_stage_anchor_generator = anchor_generator_builder.build(
frcnn_config.first_stage_anchor_generator)
first_stage_atrous_rate = frcnn_config.first_stage_atrous_rate
first_stage_box_predictor_arg_scope_fn = hyperparams_builder.build(
frcnn_config.first_stage_box_predictor_conv_hyperparams, is_training)
first_stage_box_predictor_kernel_size = (
frcnn_config.first_stage_box_predictor_kernel_size)
first_stage_box_predictor_depth = frcnn_config.first_stage_box_predictor_depth
first_stage_minibatch_size = frcnn_config.first_stage_minibatch_size
first_stage_positive_balance_fraction = (
frcnn_config.first_stage_positive_balance_fraction)
first_stage_nms_score_threshold = frcnn_config.first_stage_nms_score_threshold
first_stage_nms_iou_threshold = frcnn_config.first_stage_nms_iou_threshold
first_stage_max_proposals = frcnn_config.first_stage_max_proposals
first_stage_loc_loss_weight = (
frcnn_config.first_stage_localization_loss_weight)
first_stage_obj_loss_weight = frcnn_config.first_stage_objectness_loss_weight
initial_crop_size = frcnn_config.initial_crop_size
maxpool_kernel_size = frcnn_config.maxpool_kernel_size
maxpool_stride = frcnn_config.maxpool_stride
second_stage_box_predictor = box_predictor_builder.build(
hyperparams_builder.build,
frcnn_config.second_stage_box_predictor,
is_training=is_training,
num_classes=num_classes)
second_stage_batch_size = frcnn_config.second_stage_batch_size
second_stage_balance_fraction = frcnn_config.second_stage_balance_fraction
(second_stage_non_max_suppression_fn, second_stage_score_conversion_fn
) = post_processing_builder.build(frcnn_config.second_stage_post_processing)
second_stage_localization_loss_weight = (
frcnn_config.second_stage_localization_loss_weight)
second_stage_classification_loss = (
losses_builder.build_faster_rcnn_classification_loss(
frcnn_config.second_stage_classification_loss))
second_stage_classification_loss_weight = (
frcnn_config.second_stage_classification_loss_weight)
second_stage_mask_prediction_loss_weight = (
frcnn_config.second_stage_mask_prediction_loss_weight)
hard_example_miner = None
if frcnn_config.HasField('hard_example_miner'):
hard_example_miner = losses_builder.build_hard_example_miner(
frcnn_config.hard_example_miner,
second_stage_classification_loss_weight,
second_stage_localization_loss_weight)
common_kwargs = {
'is_training': is_training,
'num_classes': num_classes,
'image_resizer_fn': image_resizer_fn,
'feature_extractor': feature_extractor,
'number_of_stages': number_of_stages,
'first_stage_anchor_generator': first_stage_anchor_generator,
'first_stage_atrous_rate': first_stage_atrous_rate,
'first_stage_box_predictor_arg_scope_fn':
first_stage_box_predictor_arg_scope_fn,
'first_stage_box_predictor_kernel_size':
first_stage_box_predictor_kernel_size,
'first_stage_box_predictor_depth': first_stage_box_predictor_depth,
'first_stage_minibatch_size': first_stage_minibatch_size,
'first_stage_positive_balance_fraction':
first_stage_positive_balance_fraction,
'first_stage_nms_score_threshold': first_stage_nms_score_threshold,
'first_stage_nms_iou_threshold': first_stage_nms_iou_threshold,
'first_stage_max_proposals': first_stage_max_proposals,
'first_stage_localization_loss_weight': first_stage_loc_loss_weight,
'first_stage_objectness_loss_weight': first_stage_obj_loss_weight,
'second_stage_batch_size': second_stage_batch_size,
'second_stage_balance_fraction': second_stage_balance_fraction,
'second_stage_non_max_suppression_fn':
second_stage_non_max_suppression_fn,
'second_stage_score_conversion_fn': second_stage_score_conversion_fn,
'second_stage_localization_loss_weight':
second_stage_localization_loss_weight,
'second_stage_classification_loss':
second_stage_classification_loss,
'second_stage_classification_loss_weight':
second_stage_classification_loss_weight,
'hard_example_miner': hard_example_miner,
'add_summaries': add_summaries}
if isinstance(second_stage_box_predictor, box_predictor.RfcnBoxPredictor):
return rfcn_meta_arch.RFCNMetaArch(
second_stage_rfcn_box_predictor=second_stage_box_predictor,
**common_kwargs)
else:
return faster_rcnn_meta_arch.FasterRCNNMetaArch(
initial_crop_size=initial_crop_size,
maxpool_kernel_size=maxpool_kernel_size,
maxpool_stride=maxpool_stride,
second_stage_mask_rcnn_box_predictor=second_stage_box_predictor,
second_stage_mask_prediction_loss_weight=(
second_stage_mask_prediction_loss_weight),
**common_kwargs)
research/object_detection/core/box_list_ops.py
View file @ 9fce9c64
......@@ -778,7 +778,7 @@ def to_absolute_coordinates(boxlist,
height,
width,
check_range=True,
maximum_normalized_coordinate=1.01,
maximum_normalized_coordinate=1.1,
scope=None):
"""Converts normalized box coordinates to absolute pixel coordinates.
......@@ -792,7 +792,7 @@ def to_absolute_coordinates(boxlist,
width: Maximum value for width of absolute box coordinates.
check_range: If True, checks if the coordinates are normalized or not.
maximum_normalized_coordinate: Maximum coordinate value to be considered
as normalized, default to 1.01.
as normalized, default to 1.1.
scope: name scope.
Returns:
......
research/object_detection/core/box_list_ops_test.py
View file @ 9fce9c64
......@@ -931,6 +931,21 @@ class CoordinatesConversionTest(tf.test.TestCase):
out = sess.run(boxlist.get())
self.assertAllClose(out, coordinates)
def test_to_absolute_coordinates_maximum_coordinate_check(self):
coordinates = tf.constant([[0, 0, 1.2, 1.2],
[0.25, 0.25, 0.75, 0.75]], tf.float32)
img = tf.ones((128, 100, 100, 3))
boxlist = box_list.BoxList(coordinates)
absolute_boxlist = box_list_ops.to_absolute_coordinates(
boxlist,
tf.shape(img)[1],
tf.shape(img)[2],
maximum_normalized_coordinate=1.1)
with self.test_session() as sess:
with self.assertRaisesOpError('assertion failed'):
sess.run(absolute_boxlist.get())
class BoxRefinementTest(tf.test.TestCase):
......
research/object_detection/core/box_predictor.py
View file @ 9fce9c64
......@@ -79,10 +79,12 @@ class BoxPredictor(object):
Returns:
A dictionary containing at least the following tensors.
box_encodings: A list of float tensors of shape
[batch_size, num_anchors_i, q, code_size] representing the location of
the objects, where q is 1 or the number of classes. Each entry in the
list corresponds to a feature map in the input `image_features` list.
box_encodings: A list of float tensors. Each entry in the list
corresponds to a feature map in the input `image_features` list. All
tensors in the list have one of the two following shapes:
a. [batch_size, num_anchors_i, q, code_size] representing the location
of the objects, where q is 1 or the number of classes.
b. [batch_size, num_anchors_i, code_size].
class_predictions_with_background: A list of float tensors of shape
[batch_size, num_anchors_i, num_classes + 1] representing the class
predictions for the proposals. Each entry in the list corresponds to a
......@@ -120,10 +122,12 @@ class BoxPredictor(object):
Returns:
A dictionary containing at least the following tensors.
box_encodings: A list of float tensors of shape
[batch_size, num_anchors_i, q, code_size] representing the location of
the objects, where q is 1 or the number of classes. Each entry in the
list corresponds to a feature map in the input `image_features` list.
box_encodings: A list of float tensors. Each entry in the list
corresponds to a feature map in the input `image_features` list. All
tensors in the list have one of the two following shapes:
a. [batch_size, num_anchors_i, q, code_size] representing the location
of the objects, where q is 1 or the number of classes.
b. [batch_size, num_anchors_i, code_size].
class_predictions_with_background: A list of float tensors of shape
[batch_size, num_anchors_i, num_classes + 1] representing the class
predictions for the proposals. Each entry in the list corresponds to a
......@@ -765,6 +769,13 @@ class ConvolutionalBoxPredictor(BoxPredictor):
}
# TODO(rathodv): Replace with slim.arg_scope_func_key once its available
# externally.
def _arg_scope_func_key(op):
"""Returns a key that can be used to index arg_scope dictionary."""
return getattr(op, '_key_op', str(op))
# TODO(rathodv): Merge the implementation with ConvolutionalBoxPredictor above
# since they are very similar.
class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
......@@ -773,8 +784,12 @@ class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
Defines the box predictor as defined in
https://arxiv.org/abs/1708.02002. This class differs from
ConvolutionalBoxPredictor in that it shares weights and biases while
predicting from different feature maps. Separate multi-layer towers are
constructed for the box encoding and class predictors respectively.
predicting from different feature maps. However, batch_norm parameters are not
shared because the statistics of the activations vary among the different
feature maps.
Also note that separate multi-layer towers are constructed for the box
encoding and class predictors respectively.
"""
def __init__(self,
......@@ -833,14 +848,15 @@ class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
Returns:
box_encodings: A list of float tensors of shape
[batch_size, num_anchors_i, q, code_size] representing the location of
the objects, where q is 1 or the number of classes. Each entry in the
list corresponds to a feature map in the input `image_features` list.
[batch_size, num_anchors_i, code_size] representing the location of
the objects. Each entry in the list corresponds to a feature map in the
input `image_features` list.
class_predictions_with_background: A list of float tensors of shape
[batch_size, num_anchors_i, num_classes + 1] representing the class
predictions for the proposals. Each entry in the list corresponds to a
feature map in the input `image_features` list.
Raises:
ValueError: If the image feature maps do not have the same number of
channels or if the num predictions per locations is differs between the
......@@ -858,15 +874,18 @@ class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
'channels, found: {}'.format(feature_channels))
box_encodings_list = []
class_predictions_list = []
for (image_feature, num_predictions_per_location) in zip(
image_features, num_predictions_per_location_list):
for feature_index, (image_feature,
num_predictions_per_location) in enumerate(
zip(image_features,
num_predictions_per_location_list)):
# Add a slot for the background class.
with tf.variable_scope('WeightSharedConvolutionalBoxPredictor',
reuse=tf.AUTO_REUSE):
num_class_slots = self.num_classes + 1
box_encodings_net = image_feature
class_predictions_net = image_feature
with slim.arg_scope(self._conv_hyperparams_fn()):
with slim.arg_scope(self._conv_hyperparams_fn()) as sc:
apply_batch_norm = _arg_scope_func_key(slim.batch_norm) in sc
for i in range(self._num_layers_before_predictor):
box_encodings_net = slim.conv2d(
box_encodings_net,
......@@ -874,14 +893,22 @@ class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
[self._kernel_size, self._kernel_size],
stride=1,
padding='SAME',
scope='BoxEncodingPredictionTower/conv2d_{}'.format(i))
activation_fn=None,
normalizer_fn=(tf.identity if apply_batch_norm else None),
scope='BoxPredictionTower/conv2d_{}'.format(i))
if apply_batch_norm:
box_encodings_net = slim.batch_norm(
box_encodings_net,
scope='BoxPredictionTower/conv2d_{}/BatchNorm/feature_{}'.
format(i, feature_index))
box_encodings_net = tf.nn.relu6(box_encodings_net)
box_encodings = slim.conv2d(
box_encodings_net,
num_predictions_per_location * self._box_code_size,
[self._kernel_size, self._kernel_size],
activation_fn=None, stride=1, padding='SAME',
normalizer_fn=None,
scope='BoxEncodingPredictor')
scope='BoxPredictor')
for i in range(self._num_layers_before_predictor):
class_predictions_net = slim.conv2d(
......@@ -890,7 +917,15 @@ class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
[self._kernel_size, self._kernel_size],
stride=1,
padding='SAME',
activation_fn=None,
normalizer_fn=(tf.identity if apply_batch_norm else None),
scope='ClassPredictionTower/conv2d_{}'.format(i))
if apply_batch_norm:
class_predictions_net = slim.batch_norm(
class_predictions_net,
scope='ClassPredictionTower/conv2d_{}/BatchNorm/feature_{}'
.format(i, feature_index))
class_predictions_net = tf.nn.relu6(class_predictions_net)
if self._use_dropout:
class_predictions_net = slim.dropout(
class_predictions_net, keep_prob=self._dropout_keep_prob)
......@@ -912,7 +947,7 @@ class WeightSharedConvolutionalBoxPredictor(BoxPredictor):
combined_feature_map_shape[1] *
combined_feature_map_shape[2] *
num_predictions_per_location,
1, self._box_code_size]))
self._box_code_size]))
box_encodings_list.append(box_encodings)
class_predictions_with_background = tf.reshape(
class_predictions_with_background,
......
research/object_detection/core/box_predictor_test.py
View file @ 9fce9c64
......@@ -442,6 +442,24 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
text_format.Merge(conv_hyperparams_text_proto, conv_hyperparams)
return hyperparams_builder.build(conv_hyperparams, is_training=True)
def _build_conv_arg_scope_no_batch_norm(self):
conv_hyperparams = hyperparams_pb2.Hyperparams()
conv_hyperparams_text_proto = """
activation: RELU_6
regularizer {
l2_regularizer {
}
}
initializer {
random_normal_initializer {
stddev: 0.01
mean: 0.0
}
}
"""
text_format.Merge(conv_hyperparams_text_proto, conv_hyperparams)
return hyperparams_builder.build(conv_hyperparams, is_training=True)
def test_get_boxes_for_five_aspect_ratios_per_location(self):
def graph_fn(image_features):
......@@ -463,7 +481,7 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
image_features = np.random.rand(4, 8, 8, 64).astype(np.float32)
(box_encodings, objectness_predictions) = self.execute(
graph_fn, [image_features])
self.assertAllEqual(box_encodings.shape, [4, 320, 1, 4])
self.assertAllEqual(box_encodings.shape, [4, 320, 4])
self.assertAllEqual(objectness_predictions.shape, [4, 320, 1])
def test_bias_predictions_to_background_with_sigmoid_score_conversion(self):
......@@ -512,7 +530,7 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
image_features = np.random.rand(4, 8, 8, 64).astype(np.float32)
(box_encodings, class_predictions_with_background) = self.execute(
graph_fn, [image_features])
self.assertAllEqual(box_encodings.shape, [4, 320, 1, 4])
self.assertAllEqual(box_encodings.shape, [4, 320, 4])
self.assertAllEqual(class_predictions_with_background.shape,
[4, 320, num_classes_without_background+1])
......@@ -543,11 +561,12 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
image_features2 = np.random.rand(4, 8, 8, 64).astype(np.float32)
(box_encodings, class_predictions_with_background) = self.execute(
graph_fn, [image_features1, image_features2])
self.assertAllEqual(box_encodings.shape, [4, 640, 1, 4])
self.assertAllEqual(box_encodings.shape, [4, 640, 4])
self.assertAllEqual(class_predictions_with_background.shape,
[4, 640, num_classes_without_background+1])
def test_predictions_from_multiple_feature_maps_share_weights(self):
def test_predictions_from_multiple_feature_maps_share_weights_not_batchnorm(
self):
num_classes_without_background = 6
def graph_fn(image_features1, image_features2):
conv_box_predictor = box_predictor.WeightSharedConvolutionalBoxPredictor(
......@@ -574,26 +593,95 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
actual_variable_set = set(
[var.op.name for var in tf.trainable_variables()])
expected_variable_set = set([
# Box prediction tower
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxPredictionTower/conv2d_0/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxPredictionTower/conv2d_0/BatchNorm/feature_0/beta'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxEncodingPredictionTower/conv2d_0/weights'),
'BoxPredictionTower/conv2d_0/BatchNorm/feature_1/beta'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxEncodingPredictionTower/conv2d_0/BatchNorm/beta'),
'BoxPredictionTower/conv2d_1/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxEncodingPredictionTower/conv2d_1/weights'),
'BoxPredictionTower/conv2d_1/BatchNorm/feature_0/beta'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxEncodingPredictionTower/conv2d_1/BatchNorm/beta'),
'BoxPredictionTower/conv2d_1/BatchNorm/feature_1/beta'),
# Box prediction head
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxPredictor/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxPredictor/biases'),
# Class prediction tower
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_0/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_0/BatchNorm/beta'),
'ClassPredictionTower/conv2d_0/BatchNorm/feature_0/beta'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_0/BatchNorm/feature_1/beta'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_1/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_1/BatchNorm/beta'),
'ClassPredictionTower/conv2d_1/BatchNorm/feature_0/beta'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_1/BatchNorm/feature_1/beta'),
# Class prediction head
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictor/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictor/biases')])
self.assertEqual(expected_variable_set, actual_variable_set)
def test_no_batchnorm_params_when_batchnorm_is_not_configured(self):
num_classes_without_background = 6
def graph_fn(image_features1, image_features2):
conv_box_predictor = box_predictor.WeightSharedConvolutionalBoxPredictor(
is_training=False,
num_classes=num_classes_without_background,
conv_hyperparams_fn=self._build_conv_arg_scope_no_batch_norm(),
depth=32,
num_layers_before_predictor=2,
box_code_size=4)
box_predictions = conv_box_predictor.predict(
[image_features1, image_features2],
num_predictions_per_location=[5, 5],
scope='BoxPredictor')
box_encodings = tf.concat(
box_predictions[box_predictor.BOX_ENCODINGS], axis=1)
class_predictions_with_background = tf.concat(
box_predictions[box_predictor.CLASS_PREDICTIONS_WITH_BACKGROUND],
axis=1)
return (box_encodings, class_predictions_with_background)
with self.test_session(graph=tf.Graph()):
graph_fn(tf.random_uniform([4, 32, 32, 3], dtype=tf.float32),
tf.random_uniform([4, 16, 16, 3], dtype=tf.float32))
actual_variable_set = set(
[var.op.name for var in tf.trainable_variables()])
expected_variable_set = set([
# Box prediction tower
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxPredictionTower/conv2d_0/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxPredictionTower/conv2d_0/biases'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxPredictionTower/conv2d_1/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxEncodingPredictor/weights'),
'BoxPredictionTower/conv2d_1/biases'),
# Box prediction head
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxEncodingPredictor/biases'),
'BoxPredictor/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'BoxPredictor/biases'),
# Class prediction tower
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_0/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_0/biases'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_1/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictionTower/conv2d_1/biases'),
# Class prediction head
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
'ClassPredictor/weights'),
('BoxPredictor/WeightSharedConvolutionalBoxPredictor/'
......@@ -628,7 +716,7 @@ class WeightSharedConvolutionalBoxPredictorTest(test_case.TestCase):
[tf.shape(box_encodings), tf.shape(objectness_predictions)],
feed_dict={image_features:
np.random.rand(4, resolution, resolution, 64)})
self.assertAllEqual(box_encodings_shape, [4, expected_num_anchors, 1, 4])
self.assertAllEqual(box_encodings_shape, [4, expected_num_anchors, 4])
self.assertAllEqual(objectness_predictions_shape,
[4, expected_num_anchors, 1])
......
research/object_detection/core/preprocessor.py
View file @ 9fce9c64
......@@ -2128,7 +2128,8 @@ def resize_to_range(image,
max_dimension=None,
method=tf.image.ResizeMethod.BILINEAR,
align_corners=False,
pad_to_max_dimension=False):
pad_to_max_dimension=False,
per_channel_pad_value=(0, 0, 0)):
"""Resizes an image so its dimensions are within the provided value.
The output size can be described by two cases:
......@@ -2153,6 +2154,8 @@ def resize_to_range(image,
so the resulting image is of the spatial size
[max_dimension, max_dimension]. If masks are included they are padded
similarly.
per_channel_pad_value: A tuple of per-channel scalar value to use for
padding. By default pads zeros.
Returns:
Note that the position of the resized_image_shape changes based on whether
......@@ -2181,8 +2184,20 @@ def resize_to_range(image,
image, new_size[:-1], method=method, align_corners=align_corners)
if pad_to_max_dimension:
new_image = tf.image.pad_to_bounding_box(
new_image, 0, 0, max_dimension, max_dimension)
channels = tf.unstack(new_image, axis=2)
if len(channels) != len(per_channel_pad_value):
raise ValueError('Number of channels must be equal to the length of '
'per-channel pad value.')
new_image = tf.stack(
[
tf.pad(
channels[i], [[0, max_dimension - new_size[0]],
[0, max_dimension - new_size[1]]],
constant_values=per_channel_pad_value[i])
for i in range(len(channels))
],
axis=2)
new_image.set_shape([max_dimension, max_dimension, 3])
result = [new_image]
if masks is not None:
......
research/object_detection/core/preprocessor_test.py
View file @ 9fce9c64
......@@ -2316,6 +2316,46 @@ class PreprocessorTest(tf.test.TestCase):
np.random.randn(*in_shape)})
self.assertAllEqual(out_image_shape, expected_shape)
def testResizeToRangeWithPadToMaxDimensionReturnsCorrectShapes(self):
in_shape_list = [[60, 40, 3], [15, 30, 3], [15, 50, 3]]
min_dim = 50
max_dim = 100
expected_shape_list = [[100, 100, 3], [100, 100, 3], [100, 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,
pad_to_max_dimension=True)
self.assertAllEqual(out_image.shape.as_list(), expected_shape)
out_image_shape = tf.shape(out_image)
with self.test_session() as sess:
out_image_shape = sess.run(
out_image_shape, feed_dict={in_image: np.random.randn(*in_shape)})
self.assertAllEqual(out_image_shape, expected_shape)
def testResizeToRangeWithPadToMaxDimensionReturnsCorrectTensor(self):
in_image_np = np.array([[[0, 1, 2]]], np.float32)
ex_image_np = np.array(
[[[0, 1, 2], [123.68, 116.779, 103.939]],
[[123.68, 116.779, 103.939], [123.68, 116.779, 103.939]]], np.float32)
min_dim = 1
max_dim = 2
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,
pad_to_max_dimension=True,
per_channel_pad_value=(123.68, 116.779, 103.939))
with self.test_session() as sess:
out_image_np = sess.run(out_image, feed_dict={in_image: in_image_np})
self.assertAllClose(ex_image_np, out_image_np)
def testResizeToRangeWithMasksPreservesStaticSpatialShape(self):
"""Tests image resizing, checking output sizes."""
in_image_shape_list = [[60, 40, 3], [15, 30, 3]]
......
research/object_detection/core/standard_fields.py
View file @ 9fce9c64
......@@ -34,6 +34,7 @@ class InputDataFields(object):
Attributes:
image: image.
image_additional_channels: additional channels.
original_image: image in the original input size.
key: unique key corresponding to image.
source_id: source of the original image.
......@@ -66,6 +67,7 @@ class InputDataFields(object):
multiclass_scores: the label score per class for each box.
"""
image = 'image'
image_additional_channels = 'image_additional_channels'
original_image = 'original_image'
key = 'key'
source_id = 'source_id'
......@@ -161,6 +163,8 @@ class TfExampleFields(object):
height: height of image in pixels, e.g. 462
width: width of image in pixels, e.g. 581
source_id: original source of the image
image_class_text: image-level label in text format
image_class_label: image-level label in numerical format
object_class_text: labels in text format, e.g. ["person", "cat"]
object_class_label: labels in numbers, e.g. [16, 8]
object_bbox_xmin: xmin coordinates of groundtruth box, e.g. 10, 30
......@@ -195,6 +199,8 @@ class TfExampleFields(object):
height = 'image/height'
width = 'image/width'
source_id = 'image/source_id'
image_class_text = 'image/class/text'
image_class_label = 'image/class/label'
object_class_text = 'image/object/class/text'
object_class_label = 'image/object/class/label'
object_bbox_ymin = 'image/object/bbox/ymin'
......
research/object_detection/data_decoders/tf_example_decoder.py
View file @ 9fce9c64
......@@ -112,7 +112,8 @@ class TfExampleDecoder(data_decoder.DataDecoder):
label_map_proto_file=None,
use_display_name=False,
dct_method='',
num_keypoints=0):
num_keypoints=0,
num_additional_channels=0):
"""Constructor sets keys_to_features and items_to_handlers.
Args:
......@@ -133,6 +134,7 @@ class TfExampleDecoder(data_decoder.DataDecoder):
are ['INTEGER_FAST', 'INTEGER_ACCURATE']. The hint may be ignored, for
example, the jpeg library does not have that specific option.
num_keypoints: the number of keypoints per object.
num_additional_channels: how many additional channels to use.
Raises:
ValueError: If `instance_mask_type` option is not one of
......@@ -178,15 +180,28 @@ class TfExampleDecoder(data_decoder.DataDecoder):
'image/object/weight':
tf.VarLenFeature(tf.float32),
}
# We are checking `dct_method` instead of passing it directly in order to
# ensure TF version 1.6 compatibility.
if dct_method:
image = slim_example_decoder.Image(
image_key='image/encoded',
format_key='image/format',
channels=3,
dct_method=dct_method)
additional_channel_image = slim_example_decoder.Image(
image_key='image/additional_channels/encoded',
format_key='image/format',
channels=1,
repeated=True,
dct_method=dct_method)
else:
image = slim_example_decoder.Image(
image_key='image/encoded', format_key='image/format', channels=3)
additional_channel_image = slim_example_decoder.Image(
image_key='image/additional_channels/encoded',
format_key='image/format',
channels=1,
repeated=True)
self.items_to_handlers = {
fields.InputDataFields.image:
image,
......@@ -211,6 +226,13 @@ class TfExampleDecoder(data_decoder.DataDecoder):
fields.InputDataFields.groundtruth_weights: (
slim_example_decoder.Tensor('image/object/weight')),
}
if num_additional_channels > 0:
self.keys_to_features[
'image/additional_channels/encoded'] = tf.FixedLenFeature(
(num_additional_channels,), tf.string)
self.items_to_handlers[
fields.InputDataFields.
image_additional_channels] = additional_channel_image
self._num_keypoints = num_keypoints
if num_keypoints > 0:
self.keys_to_features['image/object/keypoint/x'] = (
......@@ -294,6 +316,9 @@ class TfExampleDecoder(data_decoder.DataDecoder):
[None] indicating if the boxes enclose a crowd.
Optional:
fields.InputDataFields.image_additional_channels - 3D uint8 tensor of
shape [None, None, num_additional_channels]. 1st dim is height; 2nd dim
is width; 3rd dim is the number of additional channels.
fields.InputDataFields.groundtruth_difficult - 1D bool tensor of shape
[None] indicating if the boxes represent `difficult` instances.
fields.InputDataFields.groundtruth_group_of - 1D bool tensor of shape
......@@ -316,6 +341,12 @@ class TfExampleDecoder(data_decoder.DataDecoder):
tensor_dict[fields.InputDataFields.num_groundtruth_boxes] = tf.shape(
tensor_dict[fields.InputDataFields.groundtruth_boxes])[0]
if fields.InputDataFields.image_additional_channels in tensor_dict:
channels = tensor_dict[fields.InputDataFields.image_additional_channels]
channels = tf.squeeze(channels, axis=3)
channels = tf.transpose(channels, perm=[1, 2, 0])
tensor_dict[fields.InputDataFields.image_additional_channels] = channels
def default_groundtruth_weights():
return tf.ones(
[tf.shape(tensor_dict[fields.InputDataFields.groundtruth_boxes])[0]],
......
research/object_detection/data_decoders/tf_example_decoder_test.py
View file @ 9fce9c64
......@@ -23,6 +23,7 @@ from tensorflow.core.example import example_pb2
from tensorflow.core.example import feature_pb2
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import test_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import lookup_ops
from tensorflow.python.ops import parsing_ops
......@@ -72,10 +73,41 @@ class TfExampleDecoderTest(tf.test.TestCase):
def _BytesFeatureFromList(self, ndarray):
values = ndarray.flatten().tolist()
for i in range(len(values)):
values[i] = values[i].encode('utf-8')
return feature_pb2.Feature(bytes_list=feature_pb2.BytesList(value=values))
def testDecodeAdditionalChannels(self):
image_tensor = np.random.randint(256, size=(4, 5, 3)).astype(np.uint8)
encoded_jpeg = self._EncodeImage(image_tensor)
additional_channel_tensor = np.random.randint(
256, size=(4, 5, 1)).astype(np.uint8)
encoded_additional_channel = self._EncodeImage(additional_channel_tensor)
decoded_additional_channel = self._DecodeImage(encoded_additional_channel)
example = tf.train.Example(
features=tf.train.Features(
feature={
'image/encoded':
self._BytesFeature(encoded_jpeg),
'image/additional_channels/encoded':
self._BytesFeatureFromList(
np.array([encoded_additional_channel] * 2)),
'image/format':
self._BytesFeature('jpeg'),
'image/source_id':
self._BytesFeature('image_id'),
})).SerializeToString()
example_decoder = tf_example_decoder.TfExampleDecoder(
num_additional_channels=2)
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
with self.test_session() as sess:
tensor_dict = sess.run(tensor_dict)
self.assertAllEqual(
np.concatenate([decoded_additional_channel] * 2, axis=2),
tensor_dict[fields.InputDataFields.image_additional_channels])
def testDecodeExampleWithBranchedBackupHandler(self):
example1 = example_pb2.Example(
features=feature_pb2.Features(
......@@ -304,6 +336,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
self.assertAllEqual(
2, tensor_dict[fields.InputDataFields.num_groundtruth_boxes])
@test_util.enable_c_shapes
def testDecodeKeypoint(self):
image_tensor = np.random.randint(256, size=(4, 5, 3)).astype(np.uint8)
encoded_jpeg = self._EncodeImage(image_tensor)
......@@ -331,7 +364,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
get_shape().as_list()), [None, 4])
self.assertAllEqual((tensor_dict[fields.InputDataFields.
groundtruth_keypoints].
get_shape().as_list()), [None, 3, 2])
get_shape().as_list()), [2, 3, 2])
with self.test_session() as sess:
tensor_dict = sess.run(tensor_dict)
......@@ -376,6 +409,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
self.assertAllClose(tensor_dict[fields.InputDataFields.groundtruth_weights],
np.ones(2, dtype=np.float32))
@test_util.enable_c_shapes
def testDecodeObjectLabel(self):
image_tensor = np.random.randint(256, size=(4, 5, 3)).astype(np.uint8)
encoded_jpeg = self._EncodeImage(image_tensor)
......@@ -391,7 +425,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
self.assertAllEqual((tensor_dict[
fields.InputDataFields.groundtruth_classes].get_shape().as_list()),
[None])
[2])
with self.test_session() as sess:
tensor_dict = sess.run(tensor_dict)
......@@ -522,6 +556,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
self.assertAllEqual([3, 1],
tensor_dict[fields.InputDataFields.groundtruth_classes])
@test_util.enable_c_shapes
def testDecodeObjectArea(self):
image_tensor = np.random.randint(256, size=(4, 5, 3)).astype(np.uint8)
encoded_jpeg = self._EncodeImage(image_tensor)
......@@ -536,13 +571,14 @@ class TfExampleDecoderTest(tf.test.TestCase):
tensor_dict = example_decoder.decode(tf.convert_to_tensor(example))
self.assertAllEqual((tensor_dict[fields.InputDataFields.groundtruth_area].
get_shape().as_list()), [None])
get_shape().as_list()), [2])
with self.test_session() as sess:
tensor_dict = sess.run(tensor_dict)
self.assertAllEqual(object_area,
tensor_dict[fields.InputDataFields.groundtruth_area])
@test_util.enable_c_shapes
def testDecodeObjectIsCrowd(self):
image_tensor = np.random.randint(256, size=(4, 5, 3)).astype(np.uint8)
encoded_jpeg = self._EncodeImage(image_tensor)
......@@ -558,7 +594,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
self.assertAllEqual((tensor_dict[
fields.InputDataFields.groundtruth_is_crowd].get_shape().as_list()),
[None])
[2])
with self.test_session() as sess:
tensor_dict = sess.run(tensor_dict)
......@@ -566,6 +602,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
tensor_dict[
fields.InputDataFields.groundtruth_is_crowd])
@test_util.enable_c_shapes
def testDecodeObjectDifficult(self):
image_tensor = np.random.randint(256, size=(4, 5, 3)).astype(np.uint8)
encoded_jpeg = self._EncodeImage(image_tensor)
......@@ -581,7 +618,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
self.assertAllEqual((tensor_dict[
fields.InputDataFields.groundtruth_difficult].get_shape().as_list()),
[None])
[2])
with self.test_session() as sess:
tensor_dict = sess.run(tensor_dict)
......@@ -589,6 +626,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
tensor_dict[
fields.InputDataFields.groundtruth_difficult])
@test_util.enable_c_shapes
def testDecodeObjectGroupOf(self):
image_tensor = np.random.randint(256, size=(4, 5, 3)).astype(np.uint8)
encoded_jpeg = self._EncodeImage(image_tensor)
......@@ -605,7 +643,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
self.assertAllEqual((tensor_dict[
fields.InputDataFields.groundtruth_group_of].get_shape().as_list()),
[None])
[2])
with self.test_session() as sess:
tensor_dict = sess.run(tensor_dict)
......@@ -637,6 +675,7 @@ class TfExampleDecoderTest(tf.test.TestCase):
object_weights,
tensor_dict[fields.InputDataFields.groundtruth_weights])
@test_util.enable_c_shapes
def testDecodeInstanceSegmentation(self):
num_instances = 4
image_height = 5
......@@ -673,11 +712,11 @@ class TfExampleDecoderTest(tf.test.TestCase):
self.assertAllEqual((
tensor_dict[fields.InputDataFields.groundtruth_instance_masks].
get_shape().as_list()), [None, None, None])
get_shape().as_list()), [4, 5, 3])
self.assertAllEqual((
tensor_dict[fields.InputDataFields.groundtruth_classes].
get_shape().as_list()), [None])
get_shape().as_list()), [4])
with self.test_session() as sess:
tensor_dict = sess.run(tensor_dict)
......
research/object_detection/dataset_tools/create_oid_tf_record.py
View file @ 9fce9c64
......@@ -16,7 +16,8 @@ r"""Creates TFRecords of Open Images dataset for object detection.
Example usage:
python object_detection/dataset_tools/create_oid_tf_record.py \
--input_annotations_csv=/path/to/input/annotations-human-bbox.csv \
--input_box_annotations_csv=/path/to/input/annotations-human-bbox.csv \
--input_image_label_annotations_csv=/path/to/input/annotations-label.csv \
--input_images_directory=/path/to/input/image_pixels_directory \
--input_label_map=/path/to/input/labels_bbox_545.labelmap \
--output_tf_record_path_prefix=/path/to/output/prefix.tfrecord
......@@ -27,7 +28,9 @@ https://github.com/openimages/dataset
This script will include every image found in the input_images_directory in the
output TFRecord, even if the image has no corresponding bounding box annotations
in the input_annotations_csv.
in the input_annotations_csv. If input_image_label_annotations_csv is specified,
it will add image-level labels as well. Note that the information of whether a
label is positivelly or negativelly verified is NOT added to tfrecord.
"""
from __future__ import absolute_import
from __future__ import division
......@@ -40,13 +43,16 @@ import pandas as pd
import tensorflow as tf
from object_detection.dataset_tools import oid_tfrecord_creation
from object_detection.dataset_tools import tf_record_creation_util
from object_detection.utils import label_map_util
tf.flags.DEFINE_string('input_annotations_csv', None,
tf.flags.DEFINE_string('input_box_annotations_csv', None,
'Path to CSV containing image bounding box annotations')
tf.flags.DEFINE_string('input_images_directory', None,
'Directory containing the image pixels '
'downloaded from the OpenImages GitHub repository.')
tf.flags.DEFINE_string('input_image_label_annotations_csv', None,
'Path to CSV containing image-level labels annotations')
tf.flags.DEFINE_string('input_label_map', None, 'Path to the label map proto')
tf.flags.DEFINE_string(
'output_tf_record_path_prefix', None,
......@@ -61,7 +67,7 @@ def main(_):
tf.logging.set_verbosity(tf.logging.INFO)
required_flags = [
'input_annotations_csv', 'input_images_directory', 'input_label_map',
'input_box_annotations_csv', 'input_images_directory', 'input_label_map',
'output_tf_record_path_prefix'
]
for flag_name in required_flags:
......@@ -69,17 +75,24 @@ def main(_):
raise ValueError('Flag --{} is required'.format(flag_name))
label_map = label_map_util.get_label_map_dict(FLAGS.input_label_map)
all_annotations = pd.read_csv(FLAGS.input_annotations_csv)
all_box_annotations = pd.read_csv(FLAGS.input_box_annotations_csv)
if FLAGS.input_image_label_annotations_csv:
all_label_annotations = pd.read_csv(FLAGS.input_image_label_annotations_csv)
all_label_annotations.rename(
columns={'Confidence': 'ConfidenceImageLabel'}, inplace=True)
else:
all_label_annotations = None
all_images = tf.gfile.Glob(
os.path.join(FLAGS.input_images_directory, '*.jpg'))
all_image_ids = [os.path.splitext(os.path.basename(v))[0] for v in all_images]
all_image_ids = pd.DataFrame({'ImageID': all_image_ids})
all_annotations = pd.concat([all_annotations, all_image_ids])
all_annotations = pd.concat(
[all_box_annotations, all_image_ids, all_label_annotations])
tf.logging.log(tf.logging.INFO, 'Found %d images...', len(all_image_ids))
with contextlib2.ExitStack() as tf_record_close_stack:
output_tfrecords = oid_tfrecord_creation.open_sharded_output_tfrecords(
output_tfrecords = tf_record_creation_util.open_sharded_output_tfrecords(
tf_record_close_stack, FLAGS.output_tf_record_path_prefix,
FLAGS.num_shards)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment