Spaces:
Running
Running
File size: 8,921 Bytes
0b8359d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 |
# Lint as: python2, python3
# Copyright 2020 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.
# ==============================================================================
"""Common utility for object detection tf.train.SequenceExamples."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import tensorflow.compat.v1 as tf
def context_float_feature(ndarray):
"""Converts a numpy float array to a context float feature.
Args:
ndarray: A numpy float array.
Returns:
A context float feature.
"""
feature = tf.train.Feature()
for val in ndarray:
feature.float_list.value.append(val)
return feature
def context_int64_feature(ndarray):
"""Converts a numpy array to a context int64 feature.
Args:
ndarray: A numpy int64 array.
Returns:
A context int64 feature.
"""
feature = tf.train.Feature()
for val in ndarray:
feature.int64_list.value.append(val)
return feature
def context_bytes_feature(ndarray):
"""Converts a numpy bytes array to a context bytes feature.
Args:
ndarray: A numpy bytes array.
Returns:
A context bytes feature.
"""
feature = tf.train.Feature()
for val in ndarray:
if isinstance(val, np.ndarray):
val = val.tolist()
feature.bytes_list.value.append(tf.compat.as_bytes(val))
return feature
def sequence_float_feature(ndarray):
"""Converts a numpy float array to a sequence float feature.
Args:
ndarray: A numpy float array.
Returns:
A sequence float feature.
"""
feature_list = tf.train.FeatureList()
for row in ndarray:
feature = feature_list.feature.add()
if row.size:
feature.float_list.value[:] = row
return feature_list
def sequence_int64_feature(ndarray):
"""Converts a numpy int64 array to a sequence int64 feature.
Args:
ndarray: A numpy int64 array.
Returns:
A sequence int64 feature.
"""
feature_list = tf.train.FeatureList()
for row in ndarray:
feature = feature_list.feature.add()
if row.size:
feature.int64_list.value[:] = row
return feature_list
def sequence_bytes_feature(ndarray):
"""Converts a bytes float array to a sequence bytes feature.
Args:
ndarray: A numpy bytes array.
Returns:
A sequence bytes feature.
"""
feature_list = tf.train.FeatureList()
for row in ndarray:
if isinstance(row, np.ndarray):
row = row.tolist()
feature = feature_list.feature.add()
if row:
row = [tf.compat.as_bytes(val) for val in row]
feature.bytes_list.value[:] = row
return feature_list
def boxes_to_box_components(bboxes):
"""Converts a list of numpy arrays (boxes) to box components.
Args:
bboxes: A numpy array of bounding boxes.
Returns:
Bounding box component lists.
"""
ymin_list = []
xmin_list = []
ymax_list = []
xmax_list = []
for bbox in bboxes:
bbox = np.array(bbox).astype(np.float32)
ymin, xmin, ymax, xmax = np.split(bbox, 4, axis=1)
ymin_list.append(np.reshape(ymin, [-1]))
xmin_list.append(np.reshape(xmin, [-1]))
ymax_list.append(np.reshape(ymax, [-1]))
xmax_list.append(np.reshape(xmax, [-1]))
return ymin_list, xmin_list, ymax_list, xmax_list
def make_sequence_example(dataset_name,
video_id,
encoded_images,
image_height,
image_width,
image_format=None,
image_source_ids=None,
timestamps=None,
is_annotated=None,
bboxes=None,
label_strings=None,
detection_bboxes=None,
detection_classes=None,
detection_scores=None):
"""Constructs tf.SequenceExamples.
Args:
dataset_name: String with dataset name.
video_id: String with video id.
encoded_images: A [num_frames] list (or numpy array) of encoded image
frames.
image_height: Height of the images.
image_width: Width of the images.
image_format: Format of encoded images.
image_source_ids: (Optional) A [num_frames] list of unique string ids for
each image.
timestamps: (Optional) A [num_frames] list (or numpy array) array with image
timestamps.
is_annotated: (Optional) A [num_frames] list (or numpy array) array
in which each element indicates whether the frame has been annotated
(1) or not (0).
bboxes: (Optional) A list (with num_frames elements) of [num_boxes_i, 4]
numpy float32 arrays holding boxes for each frame.
label_strings: (Optional) A list (with num_frames_elements) of [num_boxes_i]
numpy string arrays holding object string labels for each frame.
detection_bboxes: (Optional) A list (with num_frames elements) of
[num_boxes_i, 4] numpy float32 arrays holding prediction boxes for each
frame.
detection_classes: (Optional) A list (with num_frames_elements) of
[num_boxes_i] numpy int64 arrays holding predicted classes for each frame.
detection_scores: (Optional) A list (with num_frames_elements) of
[num_boxes_i] numpy float32 arrays holding predicted object scores for
each frame.
Returns:
A tf.train.SequenceExample.
"""
num_frames = len(encoded_images)
image_encoded = np.expand_dims(encoded_images, axis=-1)
if timestamps is None:
timestamps = np.arange(num_frames)
image_timestamps = np.expand_dims(timestamps, axis=-1)
# Context fields.
context_dict = {
'example/dataset_name': context_bytes_feature([dataset_name]),
'clip/start/timestamp': context_int64_feature([image_timestamps[0][0]]),
'clip/end/timestamp': context_int64_feature([image_timestamps[-1][0]]),
'clip/frames': context_int64_feature([num_frames]),
'image/channels': context_int64_feature([3]),
'image/height': context_int64_feature([image_height]),
'image/width': context_int64_feature([image_width]),
'clip/media_id': context_bytes_feature([video_id])
}
# Sequence fields.
feature_list = {
'image/encoded': sequence_bytes_feature(image_encoded),
'image/timestamp': sequence_int64_feature(image_timestamps),
}
# Add optional fields.
if image_format is not None:
context_dict['image/format'] = context_bytes_feature([image_format])
if image_source_ids is not None:
feature_list['image/source_id'] = sequence_bytes_feature(image_source_ids)
if bboxes is not None:
bbox_ymin, bbox_xmin, bbox_ymax, bbox_xmax = boxes_to_box_components(bboxes)
feature_list['region/bbox/xmin'] = sequence_float_feature(bbox_xmin)
feature_list['region/bbox/xmax'] = sequence_float_feature(bbox_xmax)
feature_list['region/bbox/ymin'] = sequence_float_feature(bbox_ymin)
feature_list['region/bbox/ymax'] = sequence_float_feature(bbox_ymax)
if is_annotated is None:
is_annotated = np.ones(num_frames, dtype=np.int64)
is_annotated = np.expand_dims(is_annotated, axis=-1)
feature_list['region/is_annotated'] = sequence_int64_feature(is_annotated)
if label_strings is not None:
feature_list['region/label/string'] = sequence_bytes_feature(
label_strings)
if detection_bboxes is not None:
det_bbox_ymin, det_bbox_xmin, det_bbox_ymax, det_bbox_xmax = (
boxes_to_box_components(detection_bboxes))
feature_list['predicted/region/bbox/xmin'] = sequence_float_feature(
det_bbox_xmin)
feature_list['predicted/region/bbox/xmax'] = sequence_float_feature(
det_bbox_xmax)
feature_list['predicted/region/bbox/ymin'] = sequence_float_feature(
det_bbox_ymin)
feature_list['predicted/region/bbox/ymax'] = sequence_float_feature(
det_bbox_ymax)
if detection_classes is not None:
feature_list['predicted/region/label/index'] = sequence_int64_feature(
detection_classes)
if detection_scores is not None:
feature_list['predicted/region/label/confidence'] = sequence_float_feature(
detection_scores)
context = tf.train.Features(feature=context_dict)
feature_lists = tf.train.FeatureLists(feature_list=feature_list)
sequence_example = tf.train.SequenceExample(
context=context,
feature_lists=feature_lists)
return sequence_example
|