def eval_image(image, height, width, bbox, thread_id, resize):
"""Get the image for model evaluation."""
with tf.name_scope('eval_image'):
if not thread_id:
tf.summary.image(
'original_image', tf.expand_dims(image, 0))
if resize == 'crop':
# Note: This is much slower than crop_to_bounding_box
# It seems that the redundant pad step has huge overhead
# distorted_image = tf.image.resize_image_with_crop_or_pad(image,
# height, width)
shape = tf.shape(image)
image = tf.cond(tf.less(shape[0], shape[1]),
lambda: tf.image.resize_images(image, tf.convert_to_tensor(
[256, 256 * shape[1] / shape[0]], dtype=tf.int32)),
lambda: tf.image.resize_images(image, tf.convert_to_tensor([256 * shape[0] / shape[1], 256], dtype=tf.int32)))
shape = tf.shape(image)
y0 = (shape[0] - height) // 2
x0 = (shape[1] - width) // 2
# y0=tf.random_uniform([],minval=0,maxval=(shape[0] - height + 1), dtype=tf.int32)
# x0=tf.random_uniform([],minval=0,maxval=(shape[1] - width + 1), dtype=tf.int32)
# distorted_image = tf.slice(image, [y0,x0,0], [height,width,3])
distorted_image = tf.image.crop_to_bounding_box(image, y0, x0, height,
width)
else:
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
tf.shape(image),
bounding_boxes=bbox,
min_object_covered=0.5,
aspect_ratio_range=[0.90, 1.10],
area_range=[0.10, 1.0],
max_attempts=100,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, _ = sample_distorted_bounding_box
# Crop the image to the specified bounding box.
distorted_image = tf.slice(image, bbox_begin, bbox_size)
resize_method = {
'nearest': tf.image.ResizeMethod.NEAREST_NEIGHBOR,
'bilinear': tf.image.ResizeMethod.BILINEAR,
'bicubic': tf.image.ResizeMethod.BICUBIC,
'area': tf.image.ResizeMethod.AREA
}[resize]
# This resizing operation may distort the images because the aspect
# ratio is not respected.
if cnn_util.tensorflow_version() >= 11:
distorted_image = tf.image.resize_images(
distorted_image, [height, width],
resize_method,
align_corners=False)
else:
distorted_image = tf.image.resize_images(
distorted_image, height, width, resize_method, align_corners=False)
distorted_image.set_shape([height, width, 3])
if not thread_id:
tf.summary.image(
'cropped_resized_image', tf.expand_dims(distorted_image, 0))
image = distorted_image
return image
def eval_image(image, height, width, bbox, thread_id, resize):
"""Get the image for model evaluation."""
with tf.name_scope('eval_image'):
if not thread_id:
tf.summary.image('original_image', tf.expand_dims(image, 0))
if resize == 'crop':
# Simple center cropping.
distorted_image = eval_image_crop(image, height, width)
elif resize == 'crop_inception':
# Eval code from Inception pre-processing.
distorted_image = eval_image_crop_inception(
image, height, width, CROP_CENTRAL_FRACTION)
elif resize == 'crop_vgg':
# Eval code from VGG pre-processing.
image = vgg_preprocessing._aspect_preserving_resize(
image, VGG_RESIZE_SIDE_MIN)
distorted_image = vgg_preprocessing._central_crop([image], height,
width)[0]
else:
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
tf.shape(image),
bounding_boxes=bbox,
min_object_covered=MIN_OBJECT_COVERED,
aspect_ratio_range=ASPECT_RATIO_RANGE,
area_range=AREA_RANGE,
max_attempts=MAX_ATTEMPTS,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, _ = sample_distorted_bounding_box
# Crop the image to the specified bounding box.
distorted_image = tf.slice(image, bbox_begin, bbox_size)
resize_method = {
'nearest': tf.image.ResizeMethod.NEAREST_NEIGHBOR,
'bilinear': tf.image.ResizeMethod.BILINEAR,
'bicubic': tf.image.ResizeMethod.BICUBIC,
'area': tf.image.ResizeMethod.AREA
}[resize]
# This resizing operation may distort the images because the aspect
# ratio is not respected.
if cnn_util.tensorflow_version() >= 11:
distorted_image = tf.image.resize_images(distorted_image,
[height, width],
resize_method,
align_corners=False)
else:
distorted_image = tf.image.resize_images(distorted_image,
height,
width,
resize_method,
align_corners=False)
distorted_image.set_shape([height, width, 3])
if not thread_id:
tf.summary.image('cropped_resized_image',
tf.expand_dims(distorted_image, 0))
image = distorted_image
return image
def eval_image(image, height, width, bbox, thread_id, resize):
"""Get the image for model evaluation."""
with tf.name_scope('eval_image'):
if not thread_id:
tf.summary.image(
'original_image', tf.expand_dims(image, 0))
if resize == 'crop':
# Note: This is much slower than crop_to_bounding_box
# It seems that the redundant pad step has huge overhead
# distorted_image = tf.image.resize_image_with_crop_or_pad(image,
# height, width)
shape = tf.shape(image)
y0 = (shape[0] - height) // 2
x0 = (shape[1] - width) // 2
# distorted_image = tf.slice(image, [y0,x0,0], [height,width,3])
distorted_image = tf.image.crop_to_bounding_box(image, y0, x0, height,
width)
else:
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
tf.shape(image),
bounding_boxes=bbox,
min_object_covered=0.1,
aspect_ratio_range=[0.75, 1.33],
area_range=[0.05, 1.0],
max_attempts=100,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, _ = sample_distorted_bounding_box
# Crop the image to the specified bounding box.
distorted_image = tf.slice(image, bbox_begin, bbox_size)
resize_method = {
'nearest': tf.image.ResizeMethod.NEAREST_NEIGHBOR,
'bilinear': tf.image.ResizeMethod.BILINEAR,
'bicubic': tf.image.ResizeMethod.BICUBIC,
'area': tf.image.ResizeMethod.AREA
}[resize]
# This resizing operation may distort the images because the aspect
# ratio is not respected.
if cnn_util.tensorflow_version() >= 11:
distorted_image = tf.image.resize_images(
distorted_image, [height, width],
resize_method,
align_corners=False)
else:
distorted_image = tf.image.resize_images(
distorted_image, height, width, resize_method, align_corners=False)
distorted_image.set_shape([height, width, 3])
if not thread_id:
tf.summary.image(
'cropped_resized_image', tf.expand_dims(distorted_image, 0))
image = distorted_image
return image
def eval_image(image, height, width, bbox, thread_id, resize_method):
"""Get the image for model evaluation."""
with tf.name_scope('eval_image'):
if not thread_id and FLAGS.summary_verbosity >= 2:
tf.summary.image('original_image', tf.expand_dims(image, 0))
if resize_method == 'crop':
# Note: This is much slower than crop_to_bounding_box
# It seems that the redundant pad step has huge overhead
# distorted_image = tf.image.resize_image_with_crop_or_pad(image,
# height, width)
shape = tf.shape(image)
y0 = (shape[0] - height) // 2
x0 = (shape[1] - width) // 2
# distorted_image = tf.slice(image, [y0,x0,0], [height,width,3])
distorted_image = tf.image.crop_to_bounding_box(
image, y0, x0, height, width)
else:
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
tf.shape(image),
bounding_boxes=bbox,
min_object_covered=0.1,
aspect_ratio_range=[0.75, 1.33],
area_range=[0.05, 1.0],
max_attempts=100,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, _ = sample_distorted_bounding_box
# Crop the image to the specified bounding box.
distorted_image = tf.slice(image, bbox_begin, bbox_size)
# TODO(reedwm): revise this resize method for eval.
image_resize_method = get_image_resize_method(
resize_method, thread_id)
# This resizing operation may distort the images because the aspect
# ratio is not respected.
if cnn_util.tensorflow_version() >= 11:
distorted_image = tf.image.resize_images(distorted_image,
[height, width],
image_resize_method,
align_corners=False)
else:
distorted_image = tf.image.resize_images(distorted_image,
height,
width,
image_resize_method,
align_corners=False)
distorted_image.set_shape([height, width, 3])
if not thread_id and FLAGS.summary_verbosity >= 2:
tf.summary.image('cropped_resized_image',
tf.expand_dims(distorted_image, 0))
image = distorted_image
return image
def train_image(image_buffer,
height,
width,
bbox,
batch_position,
resize_method,
distortions,
scope=None,
summary_verbosity=0,
distort_color_in_yiq=False,
fuse_decode_and_crop=False):
"""Distort one image for training a network.
Distorting images provides a useful technique for augmenting the data
set during training in order to make the network invariant to aspects
of the image that do not effect the label.
Args:
image_buffer: scalar string Tensor representing the raw JPEG image buffer.
height: integer
width: integer
bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
where each coordinate is [0, 1) and the coordinates are arranged
as [ymin, xmin, ymax, xmax].
batch_position: position of the image in a batch, which affects how images
are distorted and resized. NOTE: this argument can be an integer or a
tensor
resize_method: round_robin, nearest, bilinear, bicubic, or area.
distortions: If true, apply full distortions for image colors.
scope: Optional scope for op_scope.
summary_verbosity: Verbosity level for summary ops. Pass 0 to disable both
summaries and checkpoints.
distort_color_in_yiq: distort color of input images in YIQ space.
fuse_decode_and_crop: fuse the decode/crop operation.
Returns:
3-D float Tensor of distorted image used for training.
"""
# with tf.op_scope([image, height, width, bbox], scope, 'distort_image'):
# with tf.name_scope(scope, 'distort_image', [image, height, width, bbox]):
with tf.name_scope(scope or 'distort_image'):
# A large fraction of image datasets contain a human-annotated bounding box
# delineating the region of the image containing the object of interest. We
# choose to create a new bounding box for the object which is a randomly
# distorted version of the human-annotated bounding box that obeys an
# allowed range of aspect ratios, sizes and overlap with the human-annotated
# bounding box. If no box is supplied, then we assume the bounding box is
# the entire image.
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
tf.image.extract_jpeg_shape(image_buffer),
bounding_boxes=bbox,
min_object_covered=0.1,
aspect_ratio_range=[0.75, 1.33],
area_range=[0.05, 1.0],
max_attempts=100,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, distort_bbox = sample_distorted_bounding_box
if summary_verbosity >= 3:
image = tf.image.decode_jpeg(image_buffer,
channels=3,
dct_method='INTEGER_FAST')
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
image_with_distorted_box = tf.image.draw_bounding_boxes(
tf.expand_dims(image, 0), distort_bbox)
tf.summary.image('images_with_distorted_bounding_box',
image_with_distorted_box)
# Crop the image to the specified bounding box.
if fuse_decode_and_crop:
offset_y, offset_x, _ = tf.unstack(bbox_begin)
target_height, target_width, _ = tf.unstack(bbox_size)
crop_window = tf.stack(
[offset_y, offset_x, target_height, target_width])
image = tf.image.decode_and_crop_jpeg(image_buffer,
crop_window,
channels=3)
else:
image = tf.image.decode_jpeg(image_buffer,
channels=3,
dct_method='INTEGER_FAST')
image = tf.slice(image, bbox_begin, bbox_size)
if distortions:
# After this point, all image pixels reside in [0,1]. Before, they were
# uint8s in the range [0, 255].
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
# This resizing operation may distort the images because the aspect
# ratio is not respected.
image_resize_method = get_image_resize_method(resize_method,
batch_position)
if cnn_util.tensorflow_version() >= 11:
distorted_image = tf.image.resize_images(image, [height, width],
image_resize_method,
align_corners=False)
else:
distorted_image = tf.image.resize_images(image,
height,
width,
image_resize_method,
align_corners=False)
# Restore the shape since the dynamic slice based upon the bbox_size loses
# the third dimension.
distorted_image.set_shape([height, width, 3])
if summary_verbosity >= 3:
tf.summary.image('cropped_resized_image',
tf.expand_dims(distorted_image, 0))
# Randomly flip the image horizontally.
distorted_image = tf.image.random_flip_left_right(distorted_image)
if distortions:
# Randomly distort the colors.
distorted_image = distort_color(
distorted_image,
batch_position,
distort_color_in_yiq=distort_color_in_yiq)
# Note: This ensures the scaling matches the output of eval_image
distorted_image *= 255
if summary_verbosity >= 3:
tf.summary.image('final_distorted_image',
tf.expand_dims(distorted_image, 0))
return distorted_image
def distort_image(image, height, width, bbox, thread_id=0, scope=None):
"""Distort one image for training a network.
Distorting images provides a useful technique for augmenting the data
set during training in order to make the network invariant to aspects
of the image that do not effect the label.
Args:
image: 3-D float Tensor of image
height: integer
width: integer
bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
where each coordinate is [0, 1) and the coordinates are arranged
as [ymin, xmin, ymax, xmax].
thread_id: integer indicating the preprocessing thread.
scope: Optional scope for op_scope.
Returns:
3-D float Tensor of distorted image used for training.
"""
# with tf.op_scope([image, height, width, bbox], scope, 'distort_image'):
# with tf.name_scope(scope, 'distort_image', [image, height, width, bbox]):
with tf.compat.v1.name_scope(scope or 'distort_image'):
# Each bounding box has shape [1, num_boxes, box coords] and
# the coordinates are ordered [ymin, xmin, ymax, xmax].
# After this point, all image pixels reside in [0,1)
# until the very end, when they're rescaled to (-1, 1). The various
# adjust_* ops all require this range for dtype float.
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
# Display the bounding box in the first thread only.
if not thread_id:
image_with_box = tf.image.draw_bounding_boxes(tf.expand_dims(image, 0),
bbox)
tf.compat.v1.summary.image(
'image_with_bounding_boxes', image_with_box)
# A large fraction of image datasets contain a human-annotated bounding
# box delineating the region of the image containing the object of interest.
# We choose to create a new bounding box for the object which is a randomly
# distorted version of the human-annotated bounding box that obeys an allowed
# range of aspect ratios, sizes and overlap with the human-annotated
# bounding box. If no box is supplied, then we assume the bounding box is
# the entire image.
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
image_size=tf.shape(input=image),
bounding_boxes=bbox,
min_object_covered=0.1,
aspect_ratio_range=[0.99, 1.01],
area_range=[0.05, 1.0],
max_attempts=100,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, distort_bbox = sample_distorted_bounding_box
if not thread_id:
image_with_distorted_box = tf.image.draw_bounding_boxes(
tf.expand_dims(image, 0), distort_bbox)
tf.compat.v1.summary.image(
'images_with_distorted_bounding_box',
image_with_distorted_box)
# Crop the image to the specified bounding box.
distorted_image = tf.slice(image, bbox_begin, bbox_size)
# This resizing operation may distort the images because the aspect
# ratio is not respected. We select a resize method in a round robin
# fashion based on the thread number.
# Note that ResizeMethod contains 4 enumerated resizing methods.
resize_method = thread_id % 4
if cnn_util.tensorflow_version() >= 11:
distorted_image = tf.image.resize(
distorted_image, [height, width], resize_method)
else:
distorted_image = tf.image.resize(
distorted_image, height, width, resize_method)
# Restore the shape since the dynamic slice based upon the bbox_size loses
# the third dimension.
distorted_image.set_shape([height, width, 3])
if not thread_id:
tf.compat.v1.summary.image(
'cropped_resized_image',
tf.expand_dims(distorted_image, 0))
# Randomly flip the image horizontally.
distorted_image = tf.image.random_flip_left_right(distorted_image)
# Randomly distort the colors.
distorted_image = distort_color(distorted_image, thread_id)
# Note: This ensures the scaling matches the output of eval_image
distorted_image *= 256
if not thread_id:
tf.compat.v1.summary.image(
'final_distorted_image',
tf.expand_dims(distorted_image, 0))
return distorted_image
def distort_image(image, height, width, bbox, thread_id=0, scope=None):
"""Distort one image for training a network.
Distorting images provides a useful technique for augmenting the data
set during training in order to make the network invariant to aspects
of the image that do not effect the label.
Args:
image: 3-D float Tensor of image
height: integer
width: integer
bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
where each coordinate is [0, 1) and the coordinates are arranged
as [ymin, xmin, ymax, xmax].
thread_id: integer indicating the preprocessing thread.
scope: Optional scope for op_scope.
Returns:
3-D float Tensor of distorted image used for training.
"""
# with tf.op_scope([image, height, width, bbox], scope, 'distort_image'):
# with tf.name_scope(scope, 'distort_image', [image, height, width, bbox]):
with tf.name_scope(scope or 'distort_image'):
# Each bounding box has shape [1, num_boxes, box coords] and
# the coordinates are ordered [ymin, xmin, ymax, xmax].
# After this point, all image pixels reside in [0,1)
# until the very end, when they're rescaled to (-1, 1). The various
# adjust_* ops all require this range for dtype float.
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
# Display the bounding box in the first thread only.
if not thread_id:
image_with_box = tf.image.draw_bounding_boxes(tf.expand_dims(image, 0),
bbox)
tf.summary.image(
'image_with_bounding_boxes', image_with_box)
# A large fraction of image datasets contain a human-annotated bounding
# box delineating the region of the image containing the object of interest.
# We choose to create a new bounding box for the object which is a randomly
# distorted version of the human-annotated bounding box that obeys an allowed
# range of aspect ratios, sizes and overlap with the human-annotated
# bounding box. If no box is supplied, then we assume the bounding box is
# the entire image.
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
tf.shape(image),
bounding_boxes=bbox,
min_object_covered=0.1,
aspect_ratio_range=[0.75, 1.33],
area_range=[0.05, 1.0],
max_attempts=100,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, distort_bbox = sample_distorted_bounding_box
if not thread_id:
image_with_distorted_box = tf.image.draw_bounding_boxes(
tf.expand_dims(image, 0), distort_bbox)
tf.summary.image(
'images_with_distorted_bounding_box',
image_with_distorted_box)
# Crop the image to the specified bounding box.
distorted_image = tf.slice(image, bbox_begin, bbox_size)
# This resizing operation may distort the images because the aspect
# ratio is not respected. We select a resize method in a round robin
# fashion based on the thread number.
# Note that ResizeMethod contains 4 enumerated resizing methods.
resize_method = thread_id % 4
if cnn_util.tensorflow_version() >= 11:
distorted_image = tf.image.resize_images(
distorted_image, [height, width], resize_method, align_corners=False)
else:
distorted_image = tf.image.resize_images(
distorted_image, height, width, resize_method, align_corners=False)
# Restore the shape since the dynamic slice based upon the bbox_size loses
# the third dimension.
distorted_image.set_shape([height, width, 3])
if not thread_id:
tf.summary.image(
'cropped_resized_image',
tf.expand_dims(distorted_image, 0))
# Randomly flip the image horizontally.
distorted_image = tf.image.random_flip_left_right(distorted_image)
# Randomly distort the colors.
distorted_image = distort_color(distorted_image, thread_id)
# Note: This ensures the scaling matches the output of eval_image
distorted_image *= 256
if not thread_id:
tf.summary.image(
'final_distorted_image',
tf.expand_dims(distorted_image, 0))
return distorted_image
def train_image(image,
height,
width,
bbox,
batch_position,
resize_method,
distortions,
scope=None):
"""Distort one image for training a network.
Distorting images provides a useful technique for augmenting the data
set during training in order to make the network invariant to aspects
of the image that do not effect the label.
Args:
image: 3-D float Tensor of image
height: integer
width: integer
bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
where each coordinate is [0, 1) and the coordinates are arranged
as [ymin, xmin, ymax, xmax].
batch_position: position of the image in a batch, which affects how images
are distorted and resized. NOTE: this argument can be an integer or a
tensor
resize_method: round_robin, nearest, bilinear, bicubic, or area.
distortions: If true, apply full distortions for image colors.
scope: Optional scope for op_scope.
Returns:
3-D float Tensor of distorted image used for training.
"""
# with tf.op_scope([image, height, width, bbox], scope, 'distort_image'):
# with tf.name_scope(scope, 'distort_image', [image, height, width, bbox]):
with tf.name_scope(scope or 'distort_image'):
# Each bounding box has shape [1, num_boxes, box coords] and
# the coordinates are ordered [ymin, xmin, ymax, xmax].
if distortions:
# After this point, all image pixels reside in [0,1)
# until the very end, when they're rescaled to (-1, 1). The various
# adjust_* ops all require this range for dtype float.
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
# Display the bounding box.
if FLAGS.summary_verbosity >= 2:
image_with_box = tf.image.draw_bounding_boxes(
tf.expand_dims(image, 0), bbox)
tf.summary.image('image_with_bounding_boxes', image_with_box)
# A large fraction of image datasets contain a human-annotated bounding box
# delineating the region of the image containing the object of interest. We
# choose to create a new bounding box for the object which is a randomly
# distorted version of the human-annotated bounding box that obeys an
# allowed range of aspect ratios, sizes and overlap with the human-annotated
# bounding box. If no box is supplied, then we assume the bounding box is
# the entire image.
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
tf.shape(image),
bounding_boxes=bbox,
min_object_covered=0.1,
aspect_ratio_range=[0.75, 1.33],
area_range=[0.05, 1.0],
max_attempts=100,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, distort_bbox = sample_distorted_bounding_box
if FLAGS.summary_verbosity >= 2:
image_with_distorted_box = tf.image.draw_bounding_boxes(
tf.expand_dims(image, 0), distort_bbox)
tf.summary.image('images_with_distorted_bounding_box',
image_with_distorted_box)
# Crop the image to the specified bounding box.
distorted_image = tf.slice(image, bbox_begin, bbox_size)
# This resizing operation may distort the images because the aspect
# ratio is not respected.
image_resize_method = get_image_resize_method(resize_method,
batch_position)
if cnn_util.tensorflow_version() >= 11:
distorted_image = tf.image.resize_images(distorted_image,
[height, width],
image_resize_method,
align_corners=False)
else:
distorted_image = tf.image.resize_images(distorted_image,
height,
width,
image_resize_method,
align_corners=False)
# Restore the shape since the dynamic slice based upon the bbox_size loses
# the third dimension.
distorted_image.set_shape([height, width, 3])
if FLAGS.summary_verbosity >= 2:
tf.summary.image('cropped_resized_image',
tf.expand_dims(distorted_image, 0))
# Randomly flip the image horizontally.
distorted_image = tf.image.random_flip_left_right(distorted_image)
if distortions:
# Randomly distort the colors.
distorted_image = distort_color(distorted_image, batch_position)
# Note: This ensures the scaling matches the output of eval_image
distorted_image *= 256
if FLAGS.summary_verbosity >= 2:
tf.summary.image('final_distorted_image',
tf.expand_dims(distorted_image, 0))
return distorted_image
def train_image(image_buffer,
height,
width,
bbox,
batch_position,
resize_method,
distortions,
scope=None,
summary_verbosity=0,
distort_color_in_yiq=False,
fuse_decode_and_crop=False):
"""Distort one image for training a network.
Distorting images provides a useful technique for augmenting the data
set during training in order to make the network invariant to aspects
of the image that do not effect the label.
Args:
image_buffer: scalar string Tensor representing the raw JPEG image buffer.
height: integer
width: integer
bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
where each coordinate is [0, 1) and the coordinates are arranged
as [ymin, xmin, ymax, xmax].
batch_position: position of the image in a batch, which affects how images
are distorted and resized. NOTE: this argument can be an integer or a
tensor
resize_method: round_robin, nearest, bilinear, bicubic, or area.
distortions: If true, apply full distortions for image colors.
scope: Optional scope for op_scope.
summary_verbosity: Verbosity level for summary ops. Pass 0 to disable both
summaries and checkpoints.
distort_color_in_yiq: distort color of input images in YIQ space.
fuse_decode_and_crop: fuse the decode/crop operation.
Returns:
3-D float Tensor of distorted image used for training.
"""
# with tf.op_scope([image, height, width, bbox], scope, 'distort_image'):
# with tf.name_scope(scope, 'distort_image', [image, height, width, bbox]):
with tf.name_scope(scope or 'distort_image'):
# A large fraction of image datasets contain a human-annotated bounding box
# delineating the region of the image containing the object of interest. We
# choose to create a new bounding box for the object which is a randomly
# distorted version of the human-annotated bounding box that obeys an
# allowed range of aspect ratios, sizes and overlap with the human-annotated
# bounding box. If no box is supplied, then we assume the bounding box is
# the entire image.
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
tf.image.extract_jpeg_shape(image_buffer),
bounding_boxes=bbox,
min_object_covered=0.1,
aspect_ratio_range=[0.75, 1.33],
area_range=[0.05, 1.0],
max_attempts=100,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, distort_bbox = sample_distorted_bounding_box
if summary_verbosity >= 3:
image = tf.image.decode_jpeg(image_buffer, channels=3,
dct_method='INTEGER_FAST')
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
image_with_distorted_box = tf.image.draw_bounding_boxes(
tf.expand_dims(image, 0), distort_bbox)
tf.summary.image(
'images_with_distorted_bounding_box',
image_with_distorted_box)
# Crop the image to the specified bounding box.
if fuse_decode_and_crop:
offset_y, offset_x, _ = tf.unstack(bbox_begin)
target_height, target_width, _ = tf.unstack(bbox_size)
crop_window = tf.stack([offset_y, offset_x, target_height, target_width])
image = tf.image.decode_and_crop_jpeg(
image_buffer, crop_window, channels=3)
else:
image = tf.image.decode_jpeg(image_buffer, channels=3,
dct_method='INTEGER_FAST')
image = tf.slice(image, bbox_begin, bbox_size)
if distortions:
# After this point, all image pixels reside in [0,1]. Before, they were
# uint8s in the range [0, 255].
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
# This resizing operation may distort the images because the aspect
# ratio is not respected.
image_resize_method = get_image_resize_method(resize_method, batch_position)
if cnn_util.tensorflow_version() >= 11:
distorted_image = tf.image.resize_images(
image, [height, width],
image_resize_method,
align_corners=False)
else:
distorted_image = tf.image.resize_images(
image,
height,
width,
image_resize_method,
align_corners=False)
# Restore the shape since the dynamic slice based upon the bbox_size loses
# the third dimension.
distorted_image.set_shape([height, width, 3])
if summary_verbosity >= 3:
tf.summary.image(
'cropped_resized_image',
tf.expand_dims(distorted_image, 0))
# Randomly flip the image horizontally.
distorted_image = tf.image.random_flip_left_right(distorted_image)
if distortions:
# Randomly distort the colors.
distorted_image = distort_color(distorted_image, batch_position,
distort_color_in_yiq=distort_color_in_yiq)
# Note: This ensures the scaling matches the output of eval_image
distorted_image *= 255
if summary_verbosity >= 3:
tf.summary.image(
'final_distorted_image',
tf.expand_dims(distorted_image, 0))
return distorted_image
