RetinaFace训练得到的模型转换问题

[spy78111]

RetinaFace使用InsightFace的retianface训练，mxnet环境，主干模型选择resnet-50,得到训练模型后需要转换为caffe模型，在mxnet2caffe时遇到如下错误:
..
..
..
24 | face_rpn_cls_score_stride8_weight -> face_rpn_cls_score_stride8 , initialized.
face_rpn_landmark_pred_stride16_bias: (20,)->(20, 256, 1, 1)
25 | face_rpn_landmark_pred_stride16_bias -> face_rpn_landmark_pred_stride16, initialized.
('face_rpn_landmark_pred_stride16_weight', 'face_rpn_landmark_pred_stride16')
face_rpn_landmark_pred_stride16_weight: (20, 256, 1, 1)->(20, 256, 1, 1)
26 | face_rpn_landmark_pred_stride16_weight -> face_rpn_landmark_pred_stride16, initialized.
face_rpn_landmark_pred_stride32_bias: (20,)->(20, 256, 1, 1)
27 | face_rpn_landmark_pred_stride32_bias -> face_rpn_landmark_pred_stride32, initialized.
('face_rpn_landmark_pred_stride32_weight', 'face_rpn_landmark_pred_stride32')
face_rpn_landmark_pred_stride32_weight: (20, 256, 1, 1)->(20, 256, 1, 1)
28 | face_rpn_landmark_pred_stride32_weight -> face_rpn_landmark_pred_stride32, initialized.
face_rpn_landmark_pred_stride8_bias: (20,)->(20, 256, 1, 1)
29 | face_rpn_landmark_pred_stride8_bias -> face_rpn_landmark_pred_stride8, initialized.
('face_rpn_landmark_pred_stride8_weight', 'face_rpn_landmark_pred_stride8')
face_rpn_landmark_pred_stride8_weight: (20, 256, 1, 1)->(20, 256, 1, 1)
30 | face_rpn_landmark_pred_stride8_weight -> face_rpn_landmark_pred_stride8, initialized.
rf_c1_aggr_bias: (256,)->(256, 256, 3, 3)
31 | rf_c1_aggr_bias -> rf_c1_aggr , initialized.
('key in mxnet', 'rf_c1_aggr_bn_beta', True)
('key in caffe', 'rf_c1_aggr_bn_fwd_scale', False)
Traceback (most recent call last):
File "mxnet2caffe.py", line 108, in
print("{}: {}->{}".format(key_i, arg_params[key_i].shape, net.params[key_caffe][0].data.shape))
KeyError: 'rf_c1_aggr_bn_fwd_scale'
[root@TServer advanced-mxnet2caffe-master]# ('key in mxnet', 'rf_c1_aggr_bn_beta', True)
('key in caffe', 'rf_c1_aggr_bn_fwd_scale', False)
Traceback (most recent call last):
File "mxnet2caffe.py", line 108, in
print("{}: {}->{}".format(key_i, arg_params[key_i].shape, net.params[key_caffe][0].data.shape))
KeyError: 'rf_c1_aggr_bn_fwd_scale'
bash: key in mxnet,: 未找到命令...

麻烦哪位大佬帮看下什么问题，不慎感谢