def training_targets(default_anchors, class_predicts, labels):
class_predicts = nd.transpose(class_predicts, axes=(0, 2, 1))
z = MultiBoxTarget(*[default_anchors, labels, class_predicts])
box_target = z[0] # box offset target for (x, y, width, height)
box_mask = z[1] # mask is used to ignore box offsets we don't want to penalize, e.g. negative samples
cls_target = z[2] # cls_target is an array of labels for all anchors boxes
return box_target, box_mask, cls_target
def training_targets(self, default_anchors, class_predicts, labels):
'''
Helper function to obtain the bounding boxes from the anchors.
'''
class_predicts = nd.transpose(class_predicts, axes=(0, 2, 1))
box_target, box_mask, cls_target = MultiBoxTarget(default_anchors, labels, class_predicts)
return box_target, box_mask, cls_target
def training_targets(default_anchors, class_predicts, labels):
class_predicts = nd.transpose(class_predicts, axes=(0, 2, 1))
z = MultiBoxTarget(*[default_anchors, labels, class_predicts])
box_target = z[0]
box_mask = z[1] # ignore certain box offsets
cls_target = z[2]
return box_target, box_mask, cls_target
def training_targets(anchors, class_preds, labels):
'''
:param anchors: 1 x num_anchors x 4
:param class_preds: batchsize x num_anchors x num_cls
:param labels: batchsize x label_width
:return:
box_target: batchsize x (num_anchors x 4)
box_mask: batchsize x (num_anchors x 4)
cls_target: batchsize x num_anchors
'''
class_preds = class_preds.transpose(axes=(0, 2, 1)) # batchsize x num_cls x num_anchors
box_target, box_mask, cls_target = MultiBoxTarget(anchors, labels, class_preds, overlap_threshold=.5, \
ignore_label=-1, negative_mining_ratio=3, minimum_negative_samples=0, \
negative_mining_thresh=.5, variances=(0.1, 0.1, 0.2, 0.2),
name="multibox_target")
return box_target, box_mask, cls_target
def training_targets(anchors, class_preds, labels):
class_preds = class_preds.transpose(axes=(0, 2, 1))
return MultiBoxTarget(anchors, labels, class_preds)
def training_targets(anchors, class_preds, labels):
class_preds = class_preds.transpose(axes=(0, 2, 1))
return MultiBoxTarget(anchors, labels, class_preds, overlap_threshold=0.3
) # ,overlap_threshold=0.3,negative_mining_ratio=0.3