def __call__(self,
rpn_rois,
rpn_rois_num,
gt_classes,
is_crowd,
gt_boxes,
im_info,
stage=0,
max_overlap=None):
rpn_rois = rpn_rois.numpy()
rpn_rois_num = rpn_rois_num.numpy()
gt_classes = gt_classes.numpy()
gt_boxes = gt_boxes.numpy()
is_crowd = is_crowd.numpy()
im_info = im_info.numpy()
max_overlap = max_overlap if max_overlap is None else max_overlap.numpy(
)
reg_weights = [i / (stage + 1) for i in self.bbox_reg_weights]
is_cascade = True if stage > 0 else False
num_classes = 2 if is_cascade else self.num_classes
outs = generate_proposal_target(
rpn_rois, rpn_rois_num, gt_classes, is_crowd, gt_boxes, im_info,
self.batch_size_per_im, self.fg_fraction, self.fg_thresh[stage],
self.bg_thresh_hi[stage], self.bg_thresh_lo[stage], reg_weights,
num_classes, self.use_random, self.is_cls_agnostic, is_cascade,
max_overlap)
outs = [to_tensor(v) for v in outs]
for v in outs:
v.stop_gradient = True
return outs
def __call__(self,
rpn_rois,
rpn_rois_num,
gt_classes,
is_crowd,
gt_boxes,
im_info,
stage=0):
rpn_rois = rpn_rois.numpy()
rpn_rois_num = rpn_rois_num.numpy()
gt_classes = gt_classes.numpy()
gt_boxes = gt_boxes.numpy()
is_crowd = is_crowd.numpy()
im_info = im_info.numpy()
outs = generate_proposal_target(
rpn_rois, rpn_rois_num, gt_classes, is_crowd, gt_boxes, im_info,
self.batch_size_per_im, self.fg_fraction, self.fg_thresh[stage],
self.bg_thresh_hi[stage], self.bg_thresh_lo[stage],
self.bbox_reg_weights[stage], self.num_classes, self.use_random,
self.is_cls_agnostic, self.is_cascade_rcnn)
outs = [to_tensor(v) for v in outs]
for v in outs:
v.stop_gradient = True
return outs