View source on GitHub |
Greedily selects a subset of bounding boxes in descending order of score.
tf.image.non_max_suppression_with_scores(
boxes,
scores,
max_output_size,
iou_threshold=0.5,
score_threshold=float('-inf'),
soft_nms_sigma=0.0,
name=None
)
Prunes away boxes that have high intersection-over-union (IOU) overlap
with previously selected boxes. Bounding boxes are supplied as
[y1, x1, y2, x2]
, where (y1, x1)
and (y2, x2)
are the coordinates of any
diagonal pair of box corners and the coordinates can be provided as normalized
(i.e., lying in the interval [0, 1]
) or absolute. Note that this algorithm
is agnostic to where the origin is in the coordinate system. Note that this
algorithm is invariant to orthogonal transformations and translations
of the coordinate system; thus translating or reflections of the coordinate
system result in the same boxes being selected by the algorithm.
The output of this operation is a set of integers indexing into the input
collection of bounding boxes representing the selected boxes. The bounding
box coordinates corresponding to the selected indices can then be obtained
using the tf.gather
operation. For example:
selected_indices, selected_scores = tf.image.non_max_suppression_padded(
boxes, scores, max_output_size, iou_threshold=1.0, score_threshold=0.1,
soft_nms_sigma=0.5)
selected_boxes = tf.gather(boxes, selected_indices)
This function generalizes the tf.image.non_max_suppression
op by also
supporting a Soft-NMS (with Gaussian weighting) mode (c.f.
Bodla et al, https://arxiv.org/abs/1704.04503) where boxes reduce the score
of other overlapping boxes instead of directly causing them to be pruned.
Consequently, in contrast to tf.image.non_max_suppression
,
tf.image.non_max_suppression_with_scores
returns the new scores of each
input box in the second output, selected_scores
.
To enable this Soft-NMS mode, set the soft_nms_sigma
parameter to be
larger than 0. When soft_nms_sigma
equals 0, the behavior of
tf.image.non_max_suppression_with_scores
is identical to that of
tf.image.non_max_suppression
(except for the extra output) both in function
and in running time.
Note that when soft_nms_sigma
> 0, Soft-NMS is performed and iou_threshold
is ignored. iou_threshold
is only used for standard NMS.