def anchor_target_layer(rpn_cls_score, gt_boxes, gt_ishard, dontcare_areas,
im_info, _feat_stride, anchor_scales):
"""
rpn_cls_score: for pytorch (1, Ax2, H, W) bg/fg scores of previous conv layer
gt_boxes: (G, 5) vstack of [x1, y1, x2, y2, class]
gt_ishard: (G, 1), 1 or 0 indicates difficult or not
dontcare_areas: (D, 4), some areas may contains small objs but no labelling. D may be 0
im_info: a list of [image_height, image_width, scale_ratios]
_feat_stride: the downsampling ratio of feature map to the original input image
anchor_scales: the scales to the basic_anchor (basic anchor is [16, 16])
----------
Returns
----------
rpn_labels : (1, 1, HxA, W), for each anchor, 0 denotes bg, 1 fg, -1 dontcare
rpn_bbox_targets: (1, 4xA, H, W), distances of the anchors to the gt_boxes(may contains some transform)
that are the regression objectives
rpn_bbox_inside_weights: (1, 4xA, H, W) weights of each boxes, mainly accepts hyper param in cfg
rpn_bbox_outside_weights: (1, 4xA, H, W) used to balance the fg/bg,
beacuse the numbers of bgs and fgs mays significiantly different
"""
rpn_cls_score = rpn_cls_score.data.cpu().numpy()
rpn_labels, rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights = \
anchor_target_layer_py(rpn_cls_score, gt_boxes, gt_ishard, dontcare_areas, im_info, _feat_stride, anchor_scales)
rpn_labels = network.np_to_variable(rpn_labels,
is_cuda=True,
dtype=torch.LongTensor)
rpn_bbox_targets = network.np_to_variable(rpn_bbox_targets,
is_cuda=True)
rpn_bbox_inside_weights = network.np_to_variable(
rpn_bbox_inside_weights, is_cuda=True)
rpn_bbox_outside_weights = network.np_to_variable(
rpn_bbox_outside_weights, is_cuda=True)
return rpn_labels, rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights
def __init__(self, debug=False):
super(RPN, self).__init__()
self.anchor_scales = cfg.ANCHOR_SCALES
self.anchor_ratios = cfg.ANCHOR_RATIOS
self.feat_stride = cfg.FEAT_STRIDE[0]
self._vgg16 = VGG16()
self.conv1 = Conv2d(512, 512, 3, same_padding=True)
self.score_conv = Conv2d(512,
len(self.anchor_scales) *
len(self.anchor_ratios) * 2,
1,
relu=False,
same_padding=False)
self.bbox_conv = Conv2d(512,
len(self.anchor_scales) *
len(self.anchor_ratios) * 4,
1,
relu=False,
same_padding=False)
# define proposal layer
self.proposal_layer = proposal_layer_py(self.feat_stride,
self.anchor_scales,
self.anchor_ratios)
# define anchor target layer
self.anchor_target_layer = anchor_target_layer_py(
self.feat_stride, self.anchor_scales, self.anchor_ratios)
# loss
self.cross_entropy = 0
self.loss_box = 0
# for log
self.debug = debug
