def regress_by_class_rbbox(self, rois, label, bbox_pred, img_meta):
"""Regress the bbox for the predicted class. Used in Cascade R-CNN.
Args:
rois (Tensor): shape (n, 5) or (n, 6)
label (Tensor): shape (n, )
bbox_pred (Tensor): shape (n, 5*(#class+1)) or (n, 5)
img_meta (dict): Image meta info.
Returns:
Tensor: Regressed bboxes, the same shape as input rois.
"""
# import pdb
# pdb.set_trace()
assert rois.size(1) == 5 or rois.size(1) == 6
if not self.reg_class_agnostic:
label = label * 5
inds = torch.stack(
(label, label + 1, label + 2, label + 3, label + 4), 1)
bbox_pred = torch.gather(bbox_pred, 1, inds)
assert bbox_pred.size(1) == 5
if rois.size(1) == 5:
if self.with_module:
# # 原先版本
# new_rois = delta2dbbox(rois, bbox_pred, self.target_means,
# self.target_stds, img_meta['img_shape'])
# 替换为第二阶段相应的解码方案
new_rois = delta2dbbox_v2(rois, bbox_pred, self.target_means,
self.target_stds,
img_meta['img_shape'])
else:
new_rois = delta2dbbox_v3(rois, bbox_pred, self.target_means,
self.target_stds,
img_meta['img_shape'])
# choose best Rroi
new_rois = choose_best_Rroi_batch(new_rois)
else:
if self.with_module:
# bboxes = delta2dbbox(rois[:, 1:], bbox_pred, self.target_means,
# self.target_stds, img_meta['img_shape'])
# 替换为第二阶段相应的解码方案
bboxes = delta2dbbox_v2(rois[:, 1:], bbox_pred,
self.target_means, self.target_stds,
img_meta['img_shape'])
else:
bboxes = delta2dbbox_v3(rois[:, 1:], bbox_pred,
self.target_means, self.target_stds,
img_meta['img_shape'])
bboxes = choose_best_Rroi_batch(bboxes)
new_rois = torch.cat((rois[:, [0]], bboxes), dim=1)
return new_rois
def get_det_rbboxes(self,
rrois,
cls_score,
rbbox_pred,
img_shape,
scale_factor,
rescale=False,
cfg=None):
if isinstance(cls_score, list):
cls_score = sum(cls_score) / float(len(cls_score))
scores = F.softmax(cls_score, dim=1) if cls_score is not None else None
if rbbox_pred is not None:
# bboxes = delta2dbbox(rois[:, 1:], bbox_pred, self.target_means,
# self.target_stds, img_shape)
dbboxes = delta2dbbox_v2(rrois[:, 1:], rbbox_pred, self.target_means,
self.target_stds, img_shape)
else:
# bboxes = rois[:, 1:]
dbboxes = rrois[:, 1:]
# TODO: add clip here
if rescale:
# bboxes /= scale_factor
# dbboxes[:, :4] /= scale_factor
dbboxes[:, 0::5] /= scale_factor
dbboxes[:, 1::5] /= scale_factor
dbboxes[:, 2::5] /= scale_factor
dbboxes[:, 3::5] /= scale_factor
# ###########################################################
from mmdet.MARK import PRINT_RBBOX_HEAD_RS_LOSS
if PRINT_RBBOX_HEAD_RS_LOSS:
# pos_inds = pos_inds[0:min(len(pos_inds), 10)]
pred_score = scores
pred_score, pred_label = torch.max(pred_score, dim=1)
# 前景标签
pred_f_indices = pred_label != 0
if torch.sum(pred_f_indices) > 0:
print('#' * 80)
print('for pred score: ', pred_score[pred_f_indices])
print('#' * 80)
# ###########################################################
if cfg is None:
return dbboxes, scores
else:
c_device = dbboxes.device
det_bboxes, det_labels = multiclass_nms_rbbox(dbboxes, scores,
cfg.score_thr, cfg.nms,
cfg.max_per_img)
# det_bboxes = torch.from_numpy(det_bboxes).to(c_device)
# det_labels = torch.from_numpy(det_labels).to(c_device)
return det_bboxes, det_labels
def get_ori_rbboxes(self, rois, delta_rbboxes, img_shape, v2=False):
if v2:
ori_rbboxes = delta2dbbox_v2(rois[:, 1:], delta_rbboxes,
self.target_means, self.target_stds,
img_shape)
else:
ori_rbboxes = delta2dbbox_v3(rois[:, 1:], delta_rbboxes,
self.target_means, self.target_stds,
img_shape)
return ori_rbboxes
def get_det_rbboxes(self,
rrois,
cls_score,
rbbox_pred,
img_shape,
scale_factor,
rescale=False,
cfg=None):
if isinstance(cls_score, list):
cls_score = sum(cls_score) / float(len(cls_score))
scores = F.softmax(cls_score, dim=1) if cls_score is not None else None
if rbbox_pred is not None:
# bboxes = delta2dbbox(rois[:, 1:], bbox_pred, self.target_means,
# self.target_stds, img_shape)
dbboxes = delta2dbbox_v2(rrois[:,
1:], rbbox_pred, self.target_means,
self.target_stds, img_shape)
else:
# bboxes = rois[:, 1:]
dbboxes = rrois[:, 1:]
# TODO: add clip here
# dbboxes[:, 4::5] = np.pi / 2
if rescale:
# bboxes /= scale_factor
# dbboxes[:, :4] /= scale_factor
dbboxes[:, 0::5] /= scale_factor
dbboxes[:, 1::5] /= scale_factor
dbboxes[:, 2::5] /= scale_factor
dbboxes[:, 3::5] /= scale_factor
if cfg is None:
return dbboxes, scores
else:
c_device = dbboxes.device
det_bboxes, det_labels = multiclass_nms_rbbox(
dbboxes, scores, cfg.score_thr, cfg.nms, cfg.max_per_img)
# det_bboxes = torch.from_numpy(det_bboxes).to(c_device)
# det_labels = torch.from_numpy(det_labels).to(c_device)
return det_bboxes, det_labels