def test_batch_multiclass_nms_with_batch_size_1(self):
boxes = tf.constant([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]],
[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]],
tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05], [.6, 0.01], [.95, 0],
[.5, 0.01], [.3, 0.01], [.01, .85], [.01, .5]]])
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = [[[0, 10, 1, 11], [0, 0, 1, 1], [0, 999, 2, 1004],
[0, 100, 1, 101]]]
exp_nms_scores = [[.95, .9, .85, .3]]
exp_nms_classes = [[0, 0, 1, 0]]
nms_dict = post_processing.batch_multiclass_non_max_suppression(
boxes,
scores,
score_thresh,
iou_thresh,
max_size_per_class=max_output_size,
max_total_size=max_output_size)
with self.test_session() as sess:
nms_output = sess.run(nms_dict)
self.assertAllClose(nms_output['detection_boxes'], exp_nms_corners)
self.assertAllClose(nms_output['detection_scores'], exp_nms_scores)
self.assertAllClose(nms_output['detection_classes'],
exp_nms_classes)
self.assertEqual(nms_output['num_detections'], [4])
def graph_fn():
boxes = tf.constant(
[[[[0, 0, 1, 1]], [[0, 0.1, 1, 1.1]], [[0, -0.1, 1, 0.9]],
[[0, 10, 1, 11]], [[0, 10.1, 1, 11.1]], [[0, 100, 1, 101]],
[[0, 1000, 1, 1002]], [[0, 1000, 1, 1002.1]]]], tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05], [.6,
0.01], [.95, 0],
[.5, 0.01], [.3, 0.01], [.01, .85],
[.01, .5]]])
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
max_classes_per_detection = 1
use_class_agnostic_nms = True
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes,
scores,
score_thresh,
iou_thresh,
max_size_per_class=max_output_size,
max_total_size=max_output_size,
use_class_agnostic_nms=use_class_agnostic_nms,
use_static_shapes=use_static_shapes,
max_classes_per_detection=max_classes_per_detection)
self.assertIsNone(nmsed_masks)
self.assertIsNone(nmsed_additional_fields)
return (nmsed_boxes, nmsed_scores, nmsed_classes, num_detections)
def test_batch_multiclass_nms_with_masks_and_num_valid_boxes(self):
boxes = tf.constant([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]]],
[[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]],
tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05], [.6, 0.01], [.95, 0]],
[[.5, 0.01], [.3, 0.01], [.01, .85], [.01, .5]]])
masks = tf.constant([[[[[0, 1], [2, 3]], [[1, 2], [3, 4]]],
[[[2, 3], [4, 5]], [[3, 4], [5, 6]]],
[[[4, 5], [6, 7]], [[5, 6], [7, 8]]],
[[[6, 7], [8, 9]], [[7, 8], [9, 10]]]],
[[[[8, 9], [10, 11]], [[9, 10], [11, 12]]],
[[[10, 11], [12, 13]], [[11, 12], [13, 14]]],
[[[12, 13], [14, 15]], [[13, 14], [15, 16]]],
[[[14, 15], [16, 17]], [[15, 16], [17, 18]]]]],
tf.float32)
num_valid_boxes = tf.constant([1, 1], tf.int32)
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = [[[0, 0, 1, 1], [0, 0, 0, 0], [0, 0, 0, 0],
[0, 0, 0, 0]],
[[0, 10.1, 1, 11.1], [0, 0, 0, 0], [0, 0, 0, 0],
[0, 0, 0, 0]]]
exp_nms_scores = [[.9, 0, 0, 0], [.5, 0, 0, 0]]
exp_nms_classes = [[0, 0, 0, 0], [0, 0, 0, 0]]
exp_nms_masks = [[[[0, 1], [2, 3]], [[0, 0], [0, 0]], [[0, 0], [0, 0]],
[[0, 0], [0, 0]]],
[[[8, 9], [10, 11]], [[0, 0], [0, 0]], [[0, 0],
[0, 0]],
[[0, 0], [0, 0]]]]
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes,
scores,
score_thresh,
iou_thresh,
max_size_per_class=max_output_size,
max_total_size=max_output_size,
num_valid_boxes=num_valid_boxes,
masks=masks)
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
num_detections) = sess.run([
nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
num_detections
])
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
self.assertAllClose(num_detections, [1, 1])
self.assertAllClose(nmsed_masks, exp_nms_masks)
def graph_fn(boxes, scores, clip_window):
(nmsed_boxes, nmsed_scores, nmsed_classes, _, _, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size, clip_window=clip_window,
use_static_shapes=True)
return nmsed_boxes, nmsed_scores, nmsed_classes, num_detections
def _post_process(boxes, scores, additional_fields=None):
"""Applies post process to get the final detections.
Args:
boxes: A [batch_size, num_anchors, q, 4] float32 tensor containing
detections. If `q` is 1 then same boxes are used for all classes
otherwise, if `q` is equal to number of classes, class-specific boxes
are used.
scores: A [batch_size, num_anchors, num_classes] float32 tensor containing
the scores for each of the `num_anchors` detections. The scores have to be
non-negative when use_static_shapes is set True.
Returns:
num_detections: A [batch_size] int32 tensor indicating the number of
valid detections per batch item. Only the top num_detections[i] entries in
nms_boxes[i], nms_scores[i] and nms_class[i] are valid. The rest of the
entries are zero paddings.
nmsed_boxes: A [batch_size, max_detections, 4] float32 tensor
containing the non-max suppressed boxes.
nmsed_scores: A [batch_size, max_detections] float32 tensor containing
the scores for the boxes.
nmsed_classes: A [batch_size, max_detections] float32 tensor
containing the class for boxes.
"""
boxes = tf.expand_dims(boxes, axis=2)
(nmsed_boxes, nmsed_scores, nmsed_classes, _, nmsed_additional_fields,
num_detections) = batch_multiclass_non_max_suppression(
boxes,
scores,
score_thresh=options.score_thresh,
iou_thresh=options.iou_thresh,
max_size_per_class=options.max_size_per_class,
max_total_size=options.max_total_size,
additional_fields=additional_fields)
return num_detections, nmsed_boxes, nmsed_scores, nmsed_classes + 1, nmsed_additional_fields
def test_batch_multiclass_nms_with_per_image_clip_window(self):
boxes = tf.constant([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]]],
[[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]],
tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05],
[.6, 0.01], [.95, 0]],
[[.5, 0.01], [.3, 0.01],
[.01, .85], [.01, .5]]])
clip_window = tf.constant([[0., 0., 5., 5.],
[0., 0., 200., 200.]])
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = np.array([[[0, 0, 1, 1],
[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]],
[[0, 10.1, 1, 11.1],
[0, 100, 1, 101],
[0, 0, 0, 0],
[0, 0, 0, 0]]])
exp_nms_scores = np.array([[.9, 0., 0., 0.],
[.5, .3, 0, 0]])
exp_nms_classes = np.array([[0, 0, 0, 0],
[0, 0, 0, 0]])
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size, max_total_size=max_output_size,
clip_window=clip_window)
self.assertIsNone(nmsed_masks)
self.assertIsNone(nmsed_additional_fields)
# Check static shapes
self.assertAllEqual(nmsed_boxes.shape.as_list(),
exp_nms_corners.shape)
self.assertAllEqual(nmsed_scores.shape.as_list(),
exp_nms_scores.shape)
self.assertAllEqual(nmsed_classes.shape.as_list(),
exp_nms_classes.shape)
self.assertEqual(num_detections.shape.as_list(), [2])
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes,
num_detections) = sess.run([nmsed_boxes, nmsed_scores, nmsed_classes,
num_detections])
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
self.assertAllClose(num_detections, [1, 2])
def graph_fn(boxes, scores):
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size,
max_total_size=max_output_size)
self.assertIsNone(nmsed_masks)
self.assertIsNone(nmsed_additional_fields)
return (nmsed_boxes, nmsed_scores, nmsed_classes, num_detections)
def graph_fn(boxes, scores, keypoints, size, num_valid_boxes):
additional_fields = {'keypoints': keypoints, 'size': size}
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size, max_total_size=max_output_size,
num_valid_boxes=num_valid_boxes,
additional_fields=additional_fields)
self.assertIsNone(nmsed_masks)
return (nmsed_boxes, nmsed_scores, nmsed_classes,
nmsed_additional_fields['keypoints'],
nmsed_additional_fields['size'], num_detections)
def det_post_process(params, class_outputs, box_outputs, scales):
from object_detection.core.post_processing import \
batch_multiclass_non_max_suppression
cls_outputs_all, box_outputs_all = [], []
for level in range(params['min_level'], params['max_level'] + 1):
cls_outputs_all.append(
tf.reshape(class_outputs[level],
[params['batch_size'], -1, params['num_classes']]))
box_outputs_all.append(
tf.reshape(box_outputs[level], [params['batch_size'], -1, 4]))
cls_outputs_all = tf.concat(cls_outputs_all, 1)
box_outputs_all = tf.concat(box_outputs_all, 1)
probs = tf.math.sigmoid(cls_outputs_all)
# Generate location of anchors.
eval_anchors = tf.transpose(
anchors.Anchors(params['min_level'], params['max_level'],
params['num_scales'], params['aspect_ratios'],
params['anchor_scale'],
params['image_size']).boxes)
ycenter_a = (eval_anchors[0] + eval_anchors[2]) / 2
xcenter_a = (eval_anchors[1] + eval_anchors[3]) / 2
ha = eval_anchors[2] - eval_anchors[0]
wa = eval_anchors[3] - eval_anchors[1]
# Generate absolute bboxes in the units of pixels of the image.
box_outputs_per_sample = tf.transpose(box_outputs_all[0])
ty, tx, th, tw = (box_outputs_per_sample[0], box_outputs_per_sample[1],
box_outputs_per_sample[2], box_outputs_per_sample[3])
w, h = tf.math.exp(tw) * wa, tf.math.exp(th) * ha
ycenter, xcenter = ty * ha + ycenter_a, tx * wa + xcenter_a
ymin, ymax = ycenter - h / 2.0, ycenter + h / 2.0
xmin, xmax = xcenter - w / 2.0, xcenter + w / 2.0
boxes = tf.transpose(tf.stack([ymin, xmin, ymax, xmax]))
# Generate the outputs
boxes_all = tf.reshape(boxes, [params['batch_size'], -1, 1, 4])
probs_all = tf.reshape(
probs, [params['batch_size'], -1, params['num_classes']])
(boxes_tf, scores_tf, classes_tf, _, _, num_detections_tf) = \
batch_multiclass_non_max_suppression(
boxes=boxes_all, scores=probs_all, score_thresh=0.5,
iou_thresh=0.5,
max_size_per_class=anchors.MAX_DETECTIONS_PER_IMAGE,
max_total_size=anchors.MAX_DETECTIONS_PER_IMAGE,
use_combined_nms=False, use_class_agnostic_nms=True)
boxes_tf *= scales
return [boxes_tf, scores_tf, classes_tf, num_detections_tf]
def graph_fn(boxes, scores, masks):
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size, max_total_size=max_output_size,
masks=masks)
self.assertIsNone(nmsed_additional_fields)
# Check static shapes
self.assertAllEqual(nmsed_boxes.shape.as_list(), exp_nms_corners.shape)
self.assertAllEqual(nmsed_scores.shape.as_list(), exp_nms_scores.shape)
self.assertAllEqual(nmsed_classes.shape.as_list(), exp_nms_classes.shape)
self.assertAllEqual(nmsed_masks.shape.as_list(), exp_nms_masks.shape)
self.assertEqual(num_detections.shape.as_list(), [2])
return (nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
num_detections)
def test_combined_nms_with_batch_size_2(self):
"""Test use_combined_nms."""
boxes = tf.constant([[[[0, 0, 0.1, 0.1], [0, 0, 0.1, 0.1]],
[[0, 0.01, 1, 0.11], [0, 0.6, 0.1, 0.7]],
[[0, -0.01, 0.1, 0.09], [0, -0.1, 0.1, 0.09]],
[[0, 0.11, 0.1, 0.2], [0, 0.11, 0.1, 0.2]]],
[[[0, 0, 0.2, 0.2], [0, 0, 0.2, 0.2]],
[[0, 0.02, 0.2, 0.22], [0, 0.02, 0.2, 0.22]],
[[0, -0.02, 0.2, 0.19], [0, -0.02, 0.2, 0.19]],
[[0, 0.21, 0.2, 0.3], [0, 0.21, 0.2, 0.3]]]],
tf.float32)
scores = tf.constant([[[.1, 0.9], [.75, 0.8], [.6, 0.3], [0.95, 0.1]],
[[.1, 0.9], [.75, 0.8], [.6, .3], [.95, .1]]])
score_thresh = 0.1
iou_thresh = .5
max_output_size = 3
exp_nms_corners = np.array([[[0, 0.11, 0.1, 0.2], [0, 0, 0.1, 0.1],
[0, 0.6, 0.1, 0.7]],
[[0, 0.21, 0.2, 0.3], [0, 0, 0.2, 0.2],
[0, 0.02, 0.2, 0.22]]])
exp_nms_scores = np.array([[.95, .9, 0.8], [.95, .9, .75]])
exp_nms_classes = np.array([[0, 1, 1], [0, 1, 0]])
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes,
scores,
score_thresh,
iou_thresh,
max_size_per_class=max_output_size,
max_total_size=max_output_size,
use_static_shapes=True,
use_combined_nms=True)
self.assertIsNone(nmsed_masks)
self.assertIsNone(nmsed_additional_fields)
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes,
num_detections) = sess.run(
[nmsed_boxes, nmsed_scores, nmsed_classes, num_detections])
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
self.assertListEqual(num_detections.tolist(), [3, 3])
def _post_process(self,
boxes,
scores,
score_thresh=1e-6,
iou_thresh=0.5,
max_size_per_class=100,
max_total_size=300):
"""Applies post process to get the final detections.
Args:
boxes: A [batch_size, num_anchors, q, 4] float32 tensor containing
detections. If `q` is 1 then same boxes are used for all classes
otherwise, if `q` is equal to number of classes, class-specific boxes
are used.
scores: A [batch_size, num_anchors, num_classes] float32 tensor containing
the scores for each of the `num_anchors` detections. The scores have to be
non-negative when use_static_shapes is set True.
score_thresh: scalar threshold for score (low scoring boxes are removed).
iou_thresh: scalar threshold for IOU (new boxes that have high IOU overlap
with previously selected boxes are removed).
max_size_per_class: maximum number of retained boxes per class.
max_total_size: maximum number of boxes retained over all classes. By
default returns all boxes retained after capping boxes per class.
Returns:
num_detections: A [batch_size] int32 tensor indicating the number of
valid detections per batch item. Only the top num_detections[i] entries in
nms_boxes[i], nms_scores[i] and nms_class[i] are valid. The rest of the
entries are zero paddings.
nmsed_boxes: A [batch_size, max_detections, 4] float32 tensor
containing the non-max suppressed boxes.
nmsed_scores: A [batch_size, max_detections] float32 tensor containing
the scores for the boxes.
nmsed_classes: A [batch_size, max_detections] float32 tensor
containing the class for boxes.
"""
boxes = tf.expand_dims(boxes, axis=2)
(nmsed_boxes, nmsed_scores, nmsed_classes, _, _,
num_detections) = batch_multiclass_non_max_suppression(
boxes,
scores,
score_thresh=score_thresh,
iou_thresh=iou_thresh,
max_size_per_class=max_size_per_class,
max_total_size=max_total_size)
return num_detections, nmsed_boxes, nmsed_scores, nmsed_classes + 1
def test_batch_classagnostic_nms_with_batch_size_1(self,
use_static_shapes=False
):
boxes = tf.constant(
[[[[0, 0, 1, 1]], [[0, 0.1, 1, 1.1]], [[0, -0.1, 1, 0.9]],
[[0, 10, 1, 11]], [[0, 10.1, 1, 11.1]], [[0, 100, 1, 101]],
[[0, 1000, 1, 1002]], [[0, 1000, 1, 1002.1]]]], tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05], [.6, 0.01], [.95, 0],
[.5, 0.01], [.3, 0.01], [.01, .85], [.01, .5]]])
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
max_classes_per_detection = 1
use_class_agnostic_nms = True
exp_nms_corners = [[[0, 10, 1, 11], [0, 0, 1, 1], [0, 1000, 1, 1002],
[0, 100, 1, 101]]]
exp_nms_scores = [[.95, .9, .85, .3]]
exp_nms_classes = [[0, 0, 1, 0]]
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes,
scores,
score_thresh,
iou_thresh,
max_size_per_class=max_output_size,
max_total_size=max_output_size,
use_class_agnostic_nms=use_class_agnostic_nms,
use_static_shapes=use_static_shapes,
max_classes_per_detection=max_classes_per_detection)
self.assertIsNone(nmsed_masks)
self.assertIsNone(nmsed_additional_fields)
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes,
num_detections) = sess.run(
[nmsed_boxes, nmsed_scores, nmsed_classes, num_detections])
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
self.assertEqual(num_detections, [4])
def test_batch_multiclass_nms_with_batch_size_1(self):
boxes = tf.constant([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]],
[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]],
tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05],
[.6, 0.01], [.95, 0],
[.5, 0.01], [.3, 0.01],
[.01, .85], [.01, .5]]])
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = [[[0, 10, 1, 11],
[0, 0, 1, 1],
[0, 999, 2, 1004],
[0, 100, 1, 101]]]
exp_nms_scores = [[.95, .9, .85, .3]]
exp_nms_classes = [[0, 0, 1, 0]]
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size, max_total_size=max_output_size)
self.assertIsNone(nmsed_masks)
self.assertIsNone(nmsed_additional_fields)
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes,
num_detections) = sess.run([nmsed_boxes, nmsed_scores, nmsed_classes,
num_detections])
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
self.assertEqual(num_detections, [4])
def graph_fn(boxes, scores, keypoints, size):
additional_fields = {'keypoints': keypoints, 'size': size}
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size, max_total_size=max_output_size,
additional_fields=additional_fields)
self.assertIsNone(nmsed_masks)
# Check static shapes
self.assertAllEqual(nmsed_boxes.shape.as_list(), exp_nms_corners.shape)
self.assertAllEqual(nmsed_scores.shape.as_list(), exp_nms_scores.shape)
self.assertAllEqual(nmsed_classes.shape.as_list(), exp_nms_classes.shape)
self.assertEqual(len(nmsed_additional_fields),
len(exp_nms_additional_fields))
for key in exp_nms_additional_fields:
self.assertAllEqual(nmsed_additional_fields[key].shape.as_list(),
exp_nms_additional_fields[key].shape)
self.assertEqual(num_detections.shape.as_list(), [2])
return (nmsed_boxes, nmsed_scores, nmsed_classes,
nmsed_additional_fields['keypoints'],
nmsed_additional_fields['size'],
num_detections)
def test_batch_multiclass_nms_with_additional_fields_and_num_valid_boxes(
self):
boxes = tf.constant([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]]],
[[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]],
tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05],
[.6, 0.01], [.95, 0]],
[[.5, 0.01], [.3, 0.01],
[.01, .85], [.01, .5]]])
additional_fields = {
'keypoints': tf.constant(
[[[[6, 7], [8, 9]],
[[0, 1], [2, 3]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]]],
[[[13, 14], [15, 16]],
[[8, 9], [10, 11]],
[[10, 11], [12, 13]],
[[0, 0], [0, 0]]]],
tf.float32)
}
num_valid_boxes = tf.constant([1, 1], tf.int32)
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = [[[0, 0, 1, 1],
[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]],
[[0, 10.1, 1, 11.1],
[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]]]
exp_nms_scores = [[.9, 0, 0, 0],
[.5, 0, 0, 0]]
exp_nms_classes = [[0, 0, 0, 0],
[0, 0, 0, 0]]
exp_nms_additional_fields = {
'keypoints': np.array([[[[6, 7], [8, 9]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]]],
[[[13, 14], [15, 16]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]]]])
}
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size, max_total_size=max_output_size,
num_valid_boxes=num_valid_boxes,
additional_fields=additional_fields)
self.assertIsNone(nmsed_masks)
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_additional_fields,
num_detections) = sess.run([nmsed_boxes, nmsed_scores, nmsed_classes,
nmsed_additional_fields, num_detections])
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
for key in exp_nms_additional_fields:
self.assertAllClose(nmsed_additional_fields[key],
exp_nms_additional_fields[key])
self.assertAllClose(num_detections, [1, 1])
def test_batch_multiclass_nms_with_masks_and_num_valid_boxes(self):
boxes = tf.constant([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]]],
[[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]],
tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05],
[.6, 0.01], [.95, 0]],
[[.5, 0.01], [.3, 0.01],
[.01, .85], [.01, .5]]])
masks = tf.constant([[[[[0, 1], [2, 3]], [[1, 2], [3, 4]]],
[[[2, 3], [4, 5]], [[3, 4], [5, 6]]],
[[[4, 5], [6, 7]], [[5, 6], [7, 8]]],
[[[6, 7], [8, 9]], [[7, 8], [9, 10]]]],
[[[[8, 9], [10, 11]], [[9, 10], [11, 12]]],
[[[10, 11], [12, 13]], [[11, 12], [13, 14]]],
[[[12, 13], [14, 15]], [[13, 14], [15, 16]]],
[[[14, 15], [16, 17]], [[15, 16], [17, 18]]]]],
tf.float32)
num_valid_boxes = tf.constant([1, 1], tf.int32)
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = [[[0, 0, 1, 1],
[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]],
[[0, 10.1, 1, 11.1],
[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]]]
exp_nms_scores = [[.9, 0, 0, 0],
[.5, 0, 0, 0]]
exp_nms_classes = [[0, 0, 0, 0],
[0, 0, 0, 0]]
exp_nms_masks = [[[[0, 1], [2, 3]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]]],
[[[8, 9], [10, 11]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]]]]
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size, max_total_size=max_output_size,
num_valid_boxes=num_valid_boxes, masks=masks)
self.assertIsNone(nmsed_additional_fields)
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
num_detections) = sess.run([nmsed_boxes, nmsed_scores, nmsed_classes,
nmsed_masks, num_detections])
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
self.assertAllClose(num_detections, [1, 1])
self.assertAllClose(nmsed_masks, exp_nms_masks)
def test_batch_multiclass_nms_with_dynamic_batch_size(self):
boxes_placeholder = tf.placeholder(tf.float32, shape=(None, None, 2, 4))
scores_placeholder = tf.placeholder(tf.float32, shape=(None, None, 2))
masks_placeholder = tf.placeholder(tf.float32, shape=(None, None, 2, 2, 2))
boxes = np.array([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]]],
[[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]])
scores = np.array([[[.9, 0.01], [.75, 0.05],
[.6, 0.01], [.95, 0]],
[[.5, 0.01], [.3, 0.01],
[.01, .85], [.01, .5]]])
masks = np.array([[[[[0, 1], [2, 3]], [[1, 2], [3, 4]]],
[[[2, 3], [4, 5]], [[3, 4], [5, 6]]],
[[[4, 5], [6, 7]], [[5, 6], [7, 8]]],
[[[6, 7], [8, 9]], [[7, 8], [9, 10]]]],
[[[[8, 9], [10, 11]], [[9, 10], [11, 12]]],
[[[10, 11], [12, 13]], [[11, 12], [13, 14]]],
[[[12, 13], [14, 15]], [[13, 14], [15, 16]]],
[[[14, 15], [16, 17]], [[15, 16], [17, 18]]]]])
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = np.array([[[0, 10, 1, 11],
[0, 0, 1, 1],
[0, 0, 0, 0],
[0, 0, 0, 0]],
[[0, 999, 2, 1004],
[0, 10.1, 1, 11.1],
[0, 100, 1, 101],
[0, 0, 0, 0]]])
exp_nms_scores = np.array([[.95, .9, 0, 0],
[.85, .5, .3, 0]])
exp_nms_classes = np.array([[0, 0, 0, 0],
[1, 0, 0, 0]])
exp_nms_masks = np.array([[[[6, 7], [8, 9]],
[[0, 1], [2, 3]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]]],
[[[13, 14], [15, 16]],
[[8, 9], [10, 11]],
[[10, 11], [12, 13]],
[[0, 0], [0, 0]]]])
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes_placeholder, scores_placeholder, score_thresh, iou_thresh,
max_size_per_class=max_output_size, max_total_size=max_output_size,
masks=masks_placeholder)
self.assertIsNone(nmsed_additional_fields)
# Check static shapes
self.assertAllEqual(nmsed_boxes.shape.as_list(), [None, 4, 4])
self.assertAllEqual(nmsed_scores.shape.as_list(), [None, 4])
self.assertAllEqual(nmsed_classes.shape.as_list(), [None, 4])
self.assertAllEqual(nmsed_masks.shape.as_list(), [None, 4, 2, 2])
self.assertEqual(num_detections.shape.as_list(), [None])
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
num_detections) = sess.run([nmsed_boxes, nmsed_scores, nmsed_classes,
nmsed_masks, num_detections],
feed_dict={boxes_placeholder: boxes,
scores_placeholder: scores,
masks_placeholder: masks})
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
self.assertAllClose(num_detections, [2, 3])
self.assertAllClose(nmsed_masks, exp_nms_masks)
def test_batch_multiclass_nms_with_additional_fields_and_num_valid_boxes(
self):
boxes = tf.constant([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]]],
[[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]],
tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05], [.6, 0.01], [.95, 0]],
[[.5, 0.01], [.3, 0.01], [.01, .85], [.01, .5]]])
additional_fields = {
'keypoints':
tf.constant([[[[6, 7], [8, 9]], [[0, 1], [2, 3]], [[0, 0], [0, 0]],
[[0, 0], [0, 0]]],
[[[13, 14], [15, 16]], [[8, 9], [10, 11]],
[[10, 11], [12, 13]], [[0, 0], [0, 0]]]], tf.float32)
}
additional_fields['size'] = tf.constant(
[[[[7], [9]], [[1], [3]], [[0], [0]], [[0], [0]]],
[[[14], [16]], [[9], [11]], [[11], [13]], [[0], [0]]]],
tf.float32)
num_valid_boxes = tf.constant([1, 1], tf.int32)
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = [[[0, 0, 1, 1], [0, 0, 0, 0], [0, 0, 0, 0],
[0, 0, 0, 0]],
[[0, 10.1, 1, 11.1], [0, 0, 0, 0], [0, 0, 0, 0],
[0, 0, 0, 0]]]
exp_nms_scores = [[.9, 0, 0, 0], [.5, 0, 0, 0]]
exp_nms_classes = [[0, 0, 0, 0], [0, 0, 0, 0]]
exp_nms_additional_fields = {
'keypoints':
np.array([[[[6, 7], [8, 9]], [[0, 0], [0, 0]], [[0, 0], [0, 0]],
[[0, 0], [0, 0]]],
[[[13, 14], [15, 16]], [[0, 0], [0, 0]], [[0, 0], [0,
0]],
[[0, 0], [0, 0]]]])
}
exp_nms_additional_fields['size'] = np.array([[[[7], [9]], [[0], [0]],
[[0], [0]], [[0], [0]]],
[[[14], [16]],
[[0], [0]], [[0], [0]],
[[0], [0]]]])
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes,
scores,
score_thresh,
iou_thresh,
max_size_per_class=max_output_size,
max_total_size=max_output_size,
num_valid_boxes=num_valid_boxes,
additional_fields=additional_fields)
self.assertIsNone(nmsed_masks)
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_additional_fields,
num_detections) = sess.run([
nmsed_boxes, nmsed_scores, nmsed_classes,
nmsed_additional_fields, num_detections
])
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
for key in exp_nms_additional_fields:
self.assertAllClose(nmsed_additional_fields[key],
exp_nms_additional_fields[key])
self.assertAllClose(num_detections, [1, 1])
def test_batch_multiclass_nms_with_dynamic_batch_size(self):
boxes_placeholder = tf.placeholder(tf.float32,
shape=(None, None, 2, 4))
scores_placeholder = tf.placeholder(tf.float32, shape=(None, None, 2))
masks_placeholder = tf.placeholder(tf.float32,
shape=(None, None, 2, 2, 2))
boxes = np.array([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]]],
[[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]])
scores = np.array([[[.9, 0.01], [.75, 0.05], [.6, 0.01], [.95, 0]],
[[.5, 0.01], [.3, 0.01], [.01, .85], [.01, .5]]])
masks = np.array([[[[[0, 1], [2, 3]], [[1, 2], [3, 4]]],
[[[2, 3], [4, 5]], [[3, 4], [5, 6]]],
[[[4, 5], [6, 7]], [[5, 6], [7, 8]]],
[[[6, 7], [8, 9]], [[7, 8], [9, 10]]]],
[[[[8, 9], [10, 11]], [[9, 10], [11, 12]]],
[[[10, 11], [12, 13]], [[11, 12], [13, 14]]],
[[[12, 13], [14, 15]], [[13, 14], [15, 16]]],
[[[14, 15], [16, 17]], [[15, 16], [17, 18]]]]])
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = np.array([[[0, 10, 1, 11], [0, 0, 1, 1],
[0, 0, 0, 0], [0, 0, 0, 0]],
[[0, 999, 2, 1004], [0, 10.1, 1, 11.1],
[0, 100, 1, 101], [0, 0, 0, 0]]])
exp_nms_scores = np.array([[.95, .9, 0, 0], [.85, .5, .3, 0]])
exp_nms_classes = np.array([[0, 0, 0, 0], [1, 0, 0, 0]])
exp_nms_masks = np.array([[[[6, 7], [8, 9]], [[0, 1], [2, 3]],
[[0, 0], [0, 0]], [[0, 0], [0, 0]]],
[[[13, 14], [15, 16]], [[8, 9], [10, 11]],
[[10, 11], [12, 13]], [[0, 0], [0, 0]]]])
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes_placeholder,
scores_placeholder,
score_thresh,
iou_thresh,
max_size_per_class=max_output_size,
max_total_size=max_output_size,
masks=masks_placeholder)
self.assertIsNone(nmsed_additional_fields)
# Check static shapes
self.assertAllEqual(nmsed_boxes.shape.as_list(), [None, 4, 4])
self.assertAllEqual(nmsed_scores.shape.as_list(), [None, 4])
self.assertAllEqual(nmsed_classes.shape.as_list(), [None, 4])
self.assertAllEqual(nmsed_masks.shape.as_list(), [None, 4, 2, 2])
self.assertEqual(num_detections.shape.as_list(), [None])
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
num_detections) = sess.run(
[
nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
num_detections
],
feed_dict={
boxes_placeholder: boxes,
scores_placeholder: scores,
masks_placeholder: masks
})
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
self.assertAllClose(num_detections, [2, 3])
self.assertAllClose(nmsed_masks, exp_nms_masks)
def test_batch_multiclass_nms_with_additional_fields(self):
boxes = tf.constant([[[[0, 0, 1, 1], [0, 0, 4, 5]],
[[0, 0.1, 1, 1.1], [0, 0.1, 2, 1.1]],
[[0, -0.1, 1, 0.9], [0, -0.1, 1, 0.9]],
[[0, 10, 1, 11], [0, 10, 1, 11]]],
[[[0, 10.1, 1, 11.1], [0, 10.1, 1, 11.1]],
[[0, 100, 1, 101], [0, 100, 1, 101]],
[[0, 1000, 1, 1002], [0, 999, 2, 1004]],
[[0, 1000, 1, 1002.1], [0, 999, 2, 1002.7]]]],
tf.float32)
scores = tf.constant([[[.9, 0.01], [.75, 0.05],
[.6, 0.01], [.95, 0]],
[[.5, 0.01], [.3, 0.01],
[.01, .85], [.01, .5]]])
additional_fields = {
'keypoints': tf.constant(
[[[[6, 7], [8, 9]],
[[0, 1], [2, 3]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]]],
[[[13, 14], [15, 16]],
[[8, 9], [10, 11]],
[[10, 11], [12, 13]],
[[0, 0], [0, 0]]]],
tf.float32)
}
score_thresh = 0.1
iou_thresh = .5
max_output_size = 4
exp_nms_corners = np.array([[[0, 10, 1, 11],
[0, 0, 1, 1],
[0, 0, 0, 0],
[0, 0, 0, 0]],
[[0, 999, 2, 1004],
[0, 10.1, 1, 11.1],
[0, 100, 1, 101],
[0, 0, 0, 0]]])
exp_nms_scores = np.array([[.95, .9, 0, 0],
[.85, .5, .3, 0]])
exp_nms_classes = np.array([[0, 0, 0, 0],
[1, 0, 0, 0]])
exp_nms_additional_fields = {
'keypoints': np.array([[[[0, 0], [0, 0]],
[[6, 7], [8, 9]],
[[0, 0], [0, 0]],
[[0, 0], [0, 0]]],
[[[10, 11], [12, 13]],
[[13, 14], [15, 16]],
[[8, 9], [10, 11]],
[[0, 0], [0, 0]]]])
}
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_masks,
nmsed_additional_fields, num_detections
) = post_processing.batch_multiclass_non_max_suppression(
boxes, scores, score_thresh, iou_thresh,
max_size_per_class=max_output_size, max_total_size=max_output_size,
additional_fields=additional_fields)
self.assertIsNone(nmsed_masks)
# Check static shapes
self.assertAllEqual(nmsed_boxes.shape.as_list(), exp_nms_corners.shape)
self.assertAllEqual(nmsed_scores.shape.as_list(), exp_nms_scores.shape)
self.assertAllEqual(nmsed_classes.shape.as_list(), exp_nms_classes.shape)
self.assertEqual(len(nmsed_additional_fields),
len(exp_nms_additional_fields))
for key in exp_nms_additional_fields:
self.assertAllEqual(nmsed_additional_fields[key].shape.as_list(),
exp_nms_additional_fields[key].shape)
self.assertEqual(num_detections.shape.as_list(), [2])
with self.test_session() as sess:
(nmsed_boxes, nmsed_scores, nmsed_classes, nmsed_additional_fields,
num_detections) = sess.run([nmsed_boxes, nmsed_scores, nmsed_classes,
nmsed_additional_fields, num_detections])
self.assertAllClose(nmsed_boxes, exp_nms_corners)
self.assertAllClose(nmsed_scores, exp_nms_scores)
self.assertAllClose(nmsed_classes, exp_nms_classes)
for key in exp_nms_additional_fields:
self.assertAllClose(nmsed_additional_fields[key],
exp_nms_additional_fields[key])
self.assertAllClose(num_detections, [2, 3])
