def __init__(self, anchor, match_threshold=0.5, unmatched_threshold=0.5):
"""Constructs anchor labeler to assign labels to anchors.
Args:
anchor: an instance of class Anchors.
match_threshold: a float number between 0 and 1 representing the
lower-bound threshold to assign positive labels for anchors. An anchor
with a score over the threshold is labeled positive.
unmatched_threshold: a float number between 0 and 1 representing the
upper-bound threshold to assign negative labels for anchors. An anchor
with a score below the threshold is labeled negative.
"""
similarity_calc = iou_similarity.IouSimilarity()
matcher = argmax_matcher.ArgMaxMatcher(
match_threshold,
unmatched_threshold=unmatched_threshold,
negatives_lower_than_unmatched=True,
force_match_for_each_row=True)
box_coder = faster_rcnn_box_coder.FasterRcnnBoxCoder()
self._target_assigner = target_assigner.TargetAssigner(
similarity_calc, matcher, box_coder)
self._anchor = anchor
self._match_threshold = match_threshold
self._unmatched_threshold = unmatched_threshold
def __init__(self,
mix_gt_boxes: bool = True,
num_sampled_rois: int = 512,
foreground_fraction: float = 0.25,
foreground_iou_threshold: float = 0.5,
background_iou_high_threshold: float = 0.5,
background_iou_low_threshold: float = 0,
skip_subsampling: bool = False,
**kwargs):
"""Initializes a ROI sampler.
Args:
mix_gt_boxes: A `bool` of whether to mix the groundtruth boxes with
proposed ROIs.
num_sampled_rois: An `int` of the number of sampled ROIs per image.
foreground_fraction: A `float` in [0, 1], what percentage of proposed ROIs
should be sampled from the foreground boxes.
foreground_iou_threshold: A `float` that represents the IoU threshold for
a box to be considered as positive (if >= `foreground_iou_threshold`).
background_iou_high_threshold: A `float` that represents the IoU threshold
for a box to be considered as negative (if overlap in
[`background_iou_low_threshold`, `background_iou_high_threshold`]).
background_iou_low_threshold: A `float` that represents the IoU threshold
for a box to be considered as negative (if overlap in
[`background_iou_low_threshold`, `background_iou_high_threshold`])
skip_subsampling: a bool that determines if we want to skip the sampling
procedure than balances the fg/bg classes. Used for upper frcnn layers
in cascade RCNN.
**kwargs: Additional keyword arguments passed to Layer.
"""
self._config_dict = {
'mix_gt_boxes': mix_gt_boxes,
'num_sampled_rois': num_sampled_rois,
'foreground_fraction': foreground_fraction,
'foreground_iou_threshold': foreground_iou_threshold,
'background_iou_high_threshold': background_iou_high_threshold,
'background_iou_low_threshold': background_iou_low_threshold,
'skip_subsampling': skip_subsampling,
}
self._sim_calc = iou_similarity.IouSimilarity()
self._box_matcher = box_matcher.BoxMatcher(thresholds=[
background_iou_low_threshold, background_iou_high_threshold,
foreground_iou_threshold
],
indicators=[-3, -1, -2, 1])
self._target_gather = target_gather.TargetGather()
self._sampler = box_sampler.BoxSampler(num_sampled_rois,
foreground_fraction)
super(ROISampler, self).__init__(**kwargs)
def __init__(self, match_threshold=0.5, unmatched_threshold=0.5):
"""Constructs anchor labeler to assign labels to anchors.
Args:
match_threshold: a float number between 0 and 1 representing the
lower-bound threshold to assign positive labels for anchors. An anchor
with a score over the threshold is labeled positive.
unmatched_threshold: a float number between 0 and 1 representing the
upper-bound threshold to assign negative labels for anchors. An anchor
with a score below the threshold is labeled negative.
"""
self.similarity_calc = iou_similarity.IouSimilarity()
self.target_gather = target_gather.TargetGather()
self.matcher = box_matcher.BoxMatcher(
thresholds=[unmatched_threshold, match_threshold],
indicators=[-1, -2, 1],
force_match_for_each_col=True)
self.box_coder = faster_rcnn_box_coder.FasterRcnnBoxCoder()
def test_similarity_unbatched(self):
boxes = tf.constant([
[0, 0, 1, 1],
[5, 0, 10, 5],
], dtype=tf.float32)
gt_boxes = tf.constant([
[0, 0, 5, 5],
[0, 5, 5, 10],
[5, 0, 10, 5],
[5, 5, 10, 10],
],
dtype=tf.float32)
sim_calc = iou_similarity.IouSimilarity()
sim_matrix = sim_calc(boxes, gt_boxes)
self.assertAllClose(sim_matrix.numpy(),
[[0.04, 0, 0, 0], [0, 0, 1., 0]])
def __init__(self,
anchor,
match_threshold=0.7,
unmatched_threshold=0.3,
rpn_batch_size_per_im=256,
rpn_fg_fraction=0.5,
has_centerness=False,
center_match_iou_threshold=0.3,
center_unmatched_iou_threshold=0.1,
num_center_samples_per_im=256):
"""Constructs rpn anchor labeler to assign labels and centerness to anchors.
Args:
anchor: an instance of class Anchors.
match_threshold: a float number between 0 and 1 representing the
lower-bound threshold to assign positive labels for anchors. An anchor
with a score over the threshold is labeled positive.
unmatched_threshold: a float number between 0 and 1 representing the
upper-bound threshold to assign negative labels for anchors. An anchor
with a score below the threshold is labeled negative.
rpn_batch_size_per_im: number of anchors that are sampled per image.
rpn_fg_fraction:
has_centerness: whether to include centerness target creation. An anchor
is paired with one centerness score.
center_match_iou_threshold: a float number between 0 and 1 representing
the lower-bound threshold to sample foreground anchors for centerness
regression. An anchor with a score over the threshold is sampled as
foreground sample for centerness regression. We sample mostly from the
foreground region (255 out of 256 samples). That is, we sample 255 vs 1
(foreground vs background) anchor points to learn centerness regression.
center_unmatched_iou_threshold: a float number between 0 and 1
representing the lower-bound threshold to sample background anchors for
centerness regression. An anchor with a score over the threshold is
sampled as foreground sample for centerness regression. We sample very
sparsely from the background region (1 out of 256 samples). That is, we
sample 255 vs 1 (foreground vs background) anchor points to learn
centerness regression.
num_center_samples_per_im: number of anchor points per image that are
sampled as centerness targets.
"""
super(OlnAnchorLabeler,
self).__init__(anchor,
match_threshold=match_threshold,
unmatched_threshold=unmatched_threshold,
rpn_batch_size_per_im=rpn_batch_size_per_im,
rpn_fg_fraction=rpn_fg_fraction)
similarity_calc = iou_similarity.IouSimilarity()
matcher = argmax_matcher.ArgMaxMatcher(
match_threshold,
unmatched_threshold=unmatched_threshold,
negatives_lower_than_unmatched=True,
force_match_for_each_row=True)
box_coder = faster_rcnn_box_coder.FasterRcnnBoxCoder()
if has_centerness:
center_matcher = argmax_matcher.ArgMaxMatcher(
center_match_iou_threshold,
unmatched_threshold=center_match_iou_threshold,
negatives_lower_than_unmatched=True,
force_match_for_each_row=True,
)
else:
center_matcher = None
self._target_assigner = target_assigner.OlnTargetAssigner(
similarity_calc, matcher, box_coder, center_matcher=center_matcher)
self._num_center_samples_per_im = num_center_samples_per_im
self._center_unmatched_iou_threshold = center_unmatched_iou_threshold
self._rpn_batch_size_per_im = rpn_batch_size_per_im
self._rpn_fg_fraction = rpn_fg_fraction