def _patch(self, one, two):
"""Build the full 4 patch of images from sets of 2 images."""
image = tf.concat([one['image'], two['image']], axis=-3)
boxes = tf.concat([one['groundtruth_boxes'], two['groundtruth_boxes']],
axis=0)
classes = tf.concat(
[one['groundtruth_classes'], two['groundtruth_classes']], axis=0)
is_crowd = tf.concat(
[one['groundtruth_is_crowd'], two['groundtruth_is_crowd']], axis=0)
area = tf.concat([one['groundtruth_area'], two['groundtruth_area']], axis=0)
if self._mosaic_crop_mode is not None:
image, boxes, classes, is_crowd, area, _ = self._mosaic_crop_image(
image, boxes, classes, is_crowd, area)
sample = one
height, width = preprocessing_ops.get_image_shape(image)
sample['image'] = tf.cast(image, tf.uint8)
sample['groundtruth_boxes'] = boxes
sample['groundtruth_area'] = area
sample['groundtruth_classes'] = tf.cast(classes,
sample['groundtruth_classes'].dtype)
sample['groundtruth_is_crowd'] = tf.cast(is_crowd, tf.bool)
sample['width'] = tf.cast(width, sample['width'].dtype)
sample['height'] = tf.cast(height, sample['height'].dtype)
sample['num_detections'] = tf.shape(sample['groundtruth_boxes'])[1]
sample['is_mosaic'] = tf.cast(1.0, tf.bool)
del sample['shiftx']
del sample['shifty']
del sample['crop_points']
return sample
def letterbox(image: tf.Tensor,
desired_size: List[int],
letter_box: bool = True) -> Tuple[tf.Tensor, tf.Tensor]:
"""Letter box an image for image serving."""
with tf.name_scope('letter_box'):
image_size = tf.cast(preprocessing_ops.get_image_shape(image), tf.float32)
scaled_size = tf.cast(desired_size, image_size.dtype)
if letter_box:
scale = tf.minimum(scaled_size[0] / image_size[0],
scaled_size[1] / image_size[1])
scaled_size = tf.round(image_size * scale)
else:
scale = 1.0
# Computes 2D image_scale.
image_scale = scaled_size / image_size
image_offset = tf.cast((desired_size - scaled_size) * 0.5, tf.int32)
offset = (scaled_size - desired_size) * 0.5
scaled_image = tf.image.resize(
image, tf.cast(scaled_size, tf.int32), method='nearest')
output_image = tf.image.pad_to_bounding_box(scaled_image, image_offset[0],
image_offset[1],
desired_size[0],
desired_size[1])
image_info = tf.stack([
image_size,
tf.cast(desired_size, dtype=tf.float32), image_scale,
tf.cast(offset, tf.float32)
])
return output_image, image_info
def _mosaic_crop_image(self, image, boxes, classes, is_crowd, area):
"""Process a patched image in preperation for final output."""
if self._mosaic_crop_mode != 'crop':
shape = tf.cast(preprocessing_ops.get_image_shape(image),
tf.float32)
center = shape * self._mosaic_center
# shift the center of the image by applying a translation to the whole
# image
ch = tf.math.round(
preprocessing_ops.random_uniform_strong(-center[0],
center[0],
seed=self._seed))
cw = tf.math.round(
preprocessing_ops.random_uniform_strong(-center[1],
center[1],
seed=self._seed))
# clip the boxes to those with in the image
image = tfa.image.translate(image, [cw, ch],
fill_value=self._pad_value)
boxes = box_ops.denormalize_boxes(boxes, shape[:2])
boxes = boxes + tf.cast([ch, cw, ch, cw], boxes.dtype)
boxes = box_ops.clip_boxes(boxes, shape[:2])
inds = box_ops.get_non_empty_box_indices(boxes)
boxes = box_ops.normalize_boxes(boxes, shape[:2])
boxes, classes, is_crowd, area = self._select_ind(
inds,
boxes,
classes, # pylint:disable=unbalanced-tuple-unpacking
is_crowd,
area)
# warp and scale the fully stitched sample
image, _, affine = preprocessing_ops.affine_warp_image(
image, [self._output_size[0], self._output_size[1]],
scale_min=self._aug_scale_min,
scale_max=self._aug_scale_max,
translate=self._aug_rand_translate,
degrees=self._aug_rand_angle,
perspective=self._aug_rand_perspective,
random_pad=self._random_pad,
seed=self._seed)
height, width = self._output_size[0], self._output_size[1]
image = tf.image.resize(image, (height, width))
# clip and clean boxes
boxes, inds = preprocessing_ops.transform_and_clip_boxes(
boxes,
None,
affine=affine,
area_thresh=self._area_thresh,
seed=self._seed)
classes, is_crowd, area = self._select_ind(inds, classes, is_crowd,
area) # pylint:disable=unbalanced-tuple-unpacking
return image, boxes, classes, is_crowd, area, area
def _reorg_boxes(self, boxes, num_detections, image):
"""Scale and Clean boxes prior to Evaluation."""
# Build a prediciton mask to take only the number of detections
mask = tf.sequence_mask(num_detections, maxlen=tf.shape(boxes)[1])
mask = tf.cast(tf.expand_dims(mask, axis=-1), boxes.dtype)
# Denormalize the boxes by the shape of the image
inshape = tf.cast(preprocessing_ops.get_image_shape(image),
boxes.dtype)
boxes = box_ops.denormalize_boxes(boxes, inshape)
# Mask the boxes for usage
boxes *= mask
boxes += (mask - 1)
return boxes
def testGetImageShape(self, image_height, image_width): image = tf.convert_to_tensor(np.random.rand(image_height, image_width, 3)) image_shape = preprocessing_ops.get_image_shape(image) self.assertAllEqual((image_height, image_width), image_shape)
