change to_tensor to init_data

official/cv/faster_rcnn/src/FasterRcnn/fpn_neck.py

@@ -30,7 +30,7 @@ def bias_init_zeros(shape):
 def _conv(in_channels, out_channels, kernel_size=3, stride=1, padding=0, pad_mode='pad'):
     """Conv2D wrapper."""
     shape = (out_channels, in_channels, kernel_size, kernel_size)
-    weights = ms.common.initializer.initializer("XavierUniform", shape=shape, dtype=ms.float32).to_tensor()
+    weights = ms.common.initializer.initializer("XavierUniform", shape=shape, dtype=ms.float32).init_data()
     shape_bias = (out_channels,)
     biass = bias_init_zeros(shape_bias)
     return nn.Conv2d(in_channels, out_channels,

official/cv/faster_rcnn/src/FasterRcnn/rcnn.py

@@ -95,18 +95,18 @@ class Rcnn(nn.Cell):
 
         shape_0 = (self.rcnn_fc_out_channels, representation_size)
         weights_0 = ms.common.initializer.initializer("XavierUniform", shape=shape_0[::-1], \
-                                                      dtype=self.ms_type).to_tensor()
+                                                      dtype=self.ms_type).init_data()
         shape_1 = (self.rcnn_fc_out_channels, self.rcnn_fc_out_channels)
         weights_1 = ms.common.initializer.initializer("XavierUniform", shape=shape_1[::-1], \
-                                                      dtype=self.ms_type).to_tensor()
+                                                      dtype=self.ms_type).init_data()
         self.shared_fc_0 = DenseNoTranpose(representation_size, self.rcnn_fc_out_channels, weights_0)
         self.shared_fc_1 = DenseNoTranpose(self.rcnn_fc_out_channels, self.rcnn_fc_out_channels, weights_1)
 
         cls_weight = ms.common.initializer.initializer('Normal', shape=[num_classes, self.rcnn_fc_out_channels][::-1],
-                                                       dtype=self.ms_type).to_tensor()
+                                                       dtype=self.ms_type).init_data()
         reg_weight = ms.common.initializer.initializer('Normal', shape=[self.num_classes_fronted * 4,
                                                                         self.rcnn_fc_out_channels][::-1],
-                                                       dtype=self.ms_type).to_tensor()
+                                                       dtype=self.ms_type).init_data()
         self.cls_scores = DenseNoTranpose(self.rcnn_fc_out_channels, num_classes, cls_weight)
         self.reg_scores = DenseNoTranpose(self.rcnn_fc_out_channels, self.num_classes_fronted * 4, reg_weight)
 

official/cv/faster_rcnn/src/FasterRcnn/rpn.py

@@ -165,18 +165,18 @@ class RPN(nn.Cell):
 
         shp_weight_conv = (feat_channels, in_channels, 3, 3)
         shp_bias_conv = (feat_channels,)
-        weight_conv = ms.common.initializer.initializer('Normal', shape=shp_weight_conv, dtype=self.ms_type).to_tensor()
-        bias_conv = ms.common.initializer.initializer(0, shape=shp_bias_conv, dtype=self.ms_type).to_tensor()
+        weight_conv = ms.common.initializer.initializer('Normal', shape=shp_weight_conv, dtype=self.ms_type).init_data()
+        bias_conv = ms.common.initializer.initializer(0, shape=shp_bias_conv, dtype=self.ms_type).init_data()
 
         shp_weight_cls = (num_anchors * cls_out_channels, feat_channels, 1, 1)
         shp_bias_cls = (num_anchors * cls_out_channels,)
-        weight_cls = ms.common.initializer.initializer('Normal', shape=shp_weight_cls, dtype=self.ms_type).to_tensor()
-        bias_cls = ms.common.initializer.initializer(0, shape=shp_bias_cls, dtype=self.ms_type).to_tensor()
+        weight_cls = ms.common.initializer.initializer('Normal', shape=shp_weight_cls, dtype=self.ms_type).init_data()
+        bias_cls = ms.common.initializer.initializer(0, shape=shp_bias_cls, dtype=self.ms_type).init_data()
 
         shp_weight_reg = (num_anchors * 4, feat_channels, 1, 1)
         shp_bias_reg = (num_anchors * 4,)
-        weight_reg = ms.common.initializer.initializer('Normal', shape=shp_weight_reg, dtype=self.ms_type).to_tensor()
-        bias_reg = ms.common.initializer.initializer(0, shape=shp_bias_reg, dtype=self.ms_type).to_tensor()
+        weight_reg = ms.common.initializer.initializer('Normal', shape=shp_weight_reg, dtype=self.ms_type).init_data()
+        bias_reg = ms.common.initializer.initializer(0, shape=shp_bias_reg, dtype=self.ms_type).init_data()
 
         for i in range(num_layers):
             rpn_reg_cls_block = RpnRegClsBlock(in_channels, feat_channels, num_anchors, cls_out_channels, \

research/cv/CascadeRCNN/src/CascadeRcnn/fpn_neck.py

@@ -30,7 +30,7 @@ def bias_init_zeros(shape):
 def _conv(in_channels, out_channels, kernel_size=3, stride=1, padding=0, pad_mode='pad'):
     """Conv2D wrapper."""
     shape = (out_channels, in_channels, kernel_size, kernel_size)
-    weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32).to_tensor()
+    weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32).init_data()
     shape_bias = (out_channels,)
     biass = bias_init_zeros(shape_bias)
     return nn.Conv2d(in_channels, out_channels,

research/cv/CascadeRCNN/src/CascadeRcnn/rcnn.py

@@ -88,18 +88,18 @@ class Rcnn(nn.Cell):
         self.test_batch_size = cfg.test_batch_size
 
         shape_0 = (self.rcnn_fc_out_channels, representation_size)
-        weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=self.ms_type).to_tensor()
+        weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=self.ms_type).init_data()
         shape_1 = (self.rcnn_fc_out_channels, self.rcnn_fc_out_channels)
-        weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=self.ms_type).to_tensor()
+        weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=self.ms_type).init_data()
         self.shared_fc_0 = DenseNoTranpose(representation_size, self.rcnn_fc_out_channels, weights_0)
         self.shared_fc_1 = DenseNoTranpose(self.rcnn_fc_out_channels, self.rcnn_fc_out_channels, weights_1)
 
         cls_weight = initializer('Normal', shape=[num_classes, self.rcnn_fc_out_channels][::-1],
-                                 dtype=self.ms_type).to_tensor()
+                                 dtype=self.ms_type).init_data()
         reg_weight = initializer('Normal', shape=[num_classes * 4, self.rcnn_fc_out_channels][::-1],
-                                 dtype=self.ms_type).to_tensor()
+                                 dtype=self.ms_type).init_data()
         reg_weight_agn = initializer('Normal', shape=[4, self.rcnn_fc_out_channels][::-1],
-                                     dtype=mstype.float32).to_tensor()
+                                     dtype=mstype.float32).init_data()
         self.cls_scores = DenseNoTranpose(self.rcnn_fc_out_channels, num_classes, cls_weight)
         self.reg_scores = DenseNoTranpose(self.rcnn_fc_out_channels, num_classes * 4, reg_weight)
         self.reg_scores_class_ang = DenseNoTranpose(self.rcnn_fc_out_channels, 4, reg_weight_agn)

research/cv/CascadeRCNN/src/CascadeRcnn/rcnn1.py

@@ -88,18 +88,18 @@ class Rcnn_1(nn.Cell):
         self.test_batch_size = cfg.test_batch_size
 
         shape_0 = (self.rcnn_fc_out_channels, representation_size)
-        weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=self.ms_type).to_tensor()
+        weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=self.ms_type).init_data()
         shape_1 = (self.rcnn_fc_out_channels, self.rcnn_fc_out_channels)
-        weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=self.ms_type).to_tensor()
+        weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=self.ms_type).init_data()
         self.shared_fc_0 = DenseNoTranpose(representation_size, self.rcnn_fc_out_channels, weights_0)
         self.shared_fc_1 = DenseNoTranpose(self.rcnn_fc_out_channels, self.rcnn_fc_out_channels, weights_1)
 
         cls_weight = initializer('Normal', shape=[num_classes, self.rcnn_fc_out_channels][::-1],
-                                 dtype=self.ms_type).to_tensor()
+                                 dtype=self.ms_type).init_data()
         reg_weight = initializer('Normal', shape=[num_classes * 4, self.rcnn_fc_out_channels][::-1],
-                                 dtype=self.ms_type).to_tensor()
+                                 dtype=self.ms_type).init_data()
         reg_weight_agn = initializer('Normal', shape=[4, self.rcnn_fc_out_channels][::-1],
-                                     dtype=mstype.float32).to_tensor()
+                                     dtype=mstype.float32).init_data()
         self.cls_scores = DenseNoTranpose(self.rcnn_fc_out_channels, num_classes, cls_weight)
         self.reg_scores = DenseNoTranpose(self.rcnn_fc_out_channels, num_classes * 4, reg_weight)
         self.reg_scores_class_ang = DenseNoTranpose(self.rcnn_fc_out_channels, 4, reg_weight_agn)

research/cv/CascadeRCNN/src/CascadeRcnn/rcnn2.py

@@ -88,16 +88,16 @@ class Rcnn_2(nn.Cell):
         self.test_batch_size = cfg.test_batch_size
 
         shape_0 = (self.rcnn_fc_out_channels, representation_size)
-        weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=self.ms_type).to_tensor()
+        weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=self.ms_type).init_data()
         shape_1 = (self.rcnn_fc_out_channels, self.rcnn_fc_out_channels)
-        weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=self.ms_type).to_tensor()
+        weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=self.ms_type).init_data()
         self.shared_fc_0 = DenseNoTranpose(representation_size, self.rcnn_fc_out_channels, weights_0)
         self.shared_fc_1 = DenseNoTranpose(self.rcnn_fc_out_channels, self.rcnn_fc_out_channels, weights_1)
 
         cls_weight = initializer('Normal', shape=[num_classes, self.rcnn_fc_out_channels][::-1],
-                                 dtype=self.ms_type).to_tensor()
+                                 dtype=self.ms_type).init_data()
         reg_weight = initializer('Normal', shape=[num_classes * 4, self.rcnn_fc_out_channels][::-1],
-                                 dtype=self.ms_type).to_tensor()
+                                 dtype=self.ms_type).init_data()
         self.cls_scores = DenseNoTranpose(self.rcnn_fc_out_channels, num_classes, cls_weight)
         self.reg_scores = DenseNoTranpose(self.rcnn_fc_out_channels, num_classes * 4, reg_weight)
 

research/cv/CascadeRCNN/src/CascadeRcnn/rpn.py

@@ -167,18 +167,18 @@ class RPN(nn.Cell):
 
         shp_weight_conv = (feat_channels, in_channels, 3, 3)
         shp_bias_conv = (feat_channels,)
-        weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=self.ms_type).to_tensor()
-        bias_conv = initializer(0, shape=shp_bias_conv, dtype=self.ms_type).to_tensor()
+        weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=self.ms_type).init_data()
+        bias_conv = initializer(0, shape=shp_bias_conv, dtype=self.ms_type).init_data()
 
         shp_weight_cls = (num_anchors * cls_out_channels, feat_channels, 1, 1)
         shp_bias_cls = (num_anchors * cls_out_channels,)
-        weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=self.ms_type).to_tensor()
-        bias_cls = initializer(0, shape=shp_bias_cls, dtype=self.ms_type).to_tensor()
+        weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=self.ms_type).init_data()
+        bias_cls = initializer(0, shape=shp_bias_cls, dtype=self.ms_type).init_data()
 
         shp_weight_reg = (num_anchors * 4, feat_channels, 1, 1)
         shp_bias_reg = (num_anchors * 4,)
-        weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=self.ms_type).to_tensor()
-        bias_reg = initializer(0, shape=shp_bias_reg, dtype=self.ms_type).to_tensor()
+        weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=self.ms_type).init_data()
+        bias_reg = initializer(0, shape=shp_bias_reg, dtype=self.ms_type).init_data()
 
         for i in range(num_layers):
             rpn_reg_cls_block = RpnRegClsBlock(in_channels, feat_channels, num_anchors, cls_out_channels, \

research/cv/HRNetW48_cls/src/utils.py

@@ -52,7 +52,7 @@ def calculate_fan_in_and_fan_out(shape):
 def get_conv_bias(cell):
     """Bias initializer for conv."""
     weight = initializer.initializer(initializer.HeUniform(negative_slope=math.sqrt(5)),
-                                     cell.weight.shape, cell.weight.dtype).to_tensor()
+                                     cell.weight.shape, cell.weight.dtype).init_data()
     fan_in, _ = calculate_fan_in_and_fan_out(weight.shape)
     bound = 1 / math.sqrt(fan_in)
     return initializer.initializer(initializer.Uniform(scale=bound),

research/cv/HRNetW48_seg/src/seg_hrnet.py

@@ -554,7 +554,7 @@ def calculate_fan_in_and_fan_out(shape):
 def get_conv_bias(cell):
     """Bias initializer for conv."""
     weight = initializer.initializer(initializer.HeUniform(negative_slope=math.sqrt(5)),
-                                     cell.weight.shape, cell.weight.dtype).to_tensor()
+                                     cell.weight.shape, cell.weight.dtype).init_data()
     fan_in, _ = calculate_fan_in_and_fan_out(weight.shape)
     bound = 1 / math.sqrt(fan_in)
     return initializer.initializer(initializer.Uniform(scale=bound),

research/cv/OCRNet/src/utils.py

@@ -47,7 +47,7 @@ def calculate_fan_in_and_fan_out(shape):
 
 def get_conv_bias(cell):
     weight = initializer.initializer(initializer.HeUniform(negative_slope=math.sqrt(5)),
-                                     cell.weight.shape, cell.weight.dtype).to_tensor()
+                                     cell.weight.shape, cell.weight.dtype).init_data()
     fan_in, _ = calculate_fan_in_and_fan_out(weight.shape)
     bound = 1 / math.sqrt(fan_in)
     return initializer.initializer(initializer.Uniform(scale=bound),

research/cv/advanced_east/src/vgg.py

@@ -38,7 +38,7 @@ def _make_layer(base, args, batch_norm):
             weight = 'ones'
             if args.initialize_mode == "XavierUniform":
                 weight_shape = (v, in_channels, 3, 3)
-                weight = initializer('XavierUniform', shape=weight_shape, dtype=mstype.float32).to_tensor()
+                weight = initializer('XavierUniform', shape=weight_shape, dtype=mstype.float32).init_data()
 
             conv2d = nn.Conv2d(in_channels=in_channels,
                                out_channels=v,

research/cv/efficientnet-b1/src/utils.py

@@ -52,7 +52,7 @@ def calculate_fan_in_and_fan_out(shape):
 def get_conv_bias(cell):
     """Bias initializer for conv."""
     weight = initializer.initializer(initializer.HeUniform(negative_slope=math.sqrt(5)),
-                                     cell.weight.shape, cell.weight.dtype).to_tensor()
+                                     cell.weight.shape, cell.weight.dtype).init_data()
     fan_in, _ = calculate_fan_in_and_fan_out(weight.shape)
     bound = 1 / math.sqrt(fan_in)
     return initializer.initializer(initializer.Uniform(scale=bound),

research/cv/faster_rcnn_dcn/src/FasterRcnn/fpn_neck.py

@@ -30,7 +30,7 @@ def bias_init_zeros(shape):
 def _conv(in_channels, out_channels, kernel_size=3, stride=1, padding=0, pad_mode='pad'):
     """Conv2D wrapper."""
     shape = (out_channels, in_channels, kernel_size, kernel_size)
-    weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32).to_tensor()
+    weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32).init_data()
     shape_bias = (out_channels,)
     biass = bias_init_zeros(shape_bias)
     return nn.Conv2d(in_channels, out_channels,

research/cv/faster_rcnn_dcn/src/FasterRcnn/rcnn.py

@@ -88,16 +88,16 @@ class Rcnn(nn.Cell):
         self.test_batch_size = cfg.test_batch_size
 
         shape_0 = (self.rcnn_fc_out_channels, representation_size)
-        weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=self.ms_type).to_tensor()
+        weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=self.ms_type).init_data()
         shape_1 = (self.rcnn_fc_out_channels, self.rcnn_fc_out_channels)
-        weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=self.ms_type).to_tensor()
+        weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=self.ms_type).init_data()
         self.shared_fc_0 = DenseNoTranpose(representation_size, self.rcnn_fc_out_channels, weights_0)
         self.shared_fc_1 = DenseNoTranpose(self.rcnn_fc_out_channels, self.rcnn_fc_out_channels, weights_1)
 
         cls_weight = initializer('Normal', shape=[num_classes, self.rcnn_fc_out_channels][::-1],
-                                 dtype=self.ms_type).to_tensor()
+                                 dtype=self.ms_type).init_data()
         reg_weight = initializer('Normal', shape=[num_classes * 4, self.rcnn_fc_out_channels][::-1],
-                                 dtype=self.ms_type).to_tensor()
+                                 dtype=self.ms_type).init_data()
         self.cls_scores = DenseNoTranpose(self.rcnn_fc_out_channels, num_classes, cls_weight)
         self.reg_scores = DenseNoTranpose(self.rcnn_fc_out_channels, num_classes * 4, reg_weight)
 

research/cv/faster_rcnn_dcn/src/FasterRcnn/rpn.py

@@ -168,18 +168,18 @@ class RPN(nn.Cell):
 
         shp_weight_conv = (feat_channels, in_channels, 3, 3)
         shp_bias_conv = (feat_channels,)
-        weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=self.ms_type).to_tensor()
-        bias_conv = initializer(0, shape=shp_bias_conv, dtype=self.ms_type).to_tensor()
+        weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=self.ms_type).init_data()
+        bias_conv = initializer(0, shape=shp_bias_conv, dtype=self.ms_type).init_data()
 
         shp_weight_cls = (num_anchors * cls_out_channels, feat_channels, 1, 1)
         shp_bias_cls = (num_anchors * cls_out_channels,)
-        weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=self.ms_type).to_tensor()
-        bias_cls = initializer(0, shape=shp_bias_cls, dtype=self.ms_type).to_tensor()
+        weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=self.ms_type).init_data()
+        bias_cls = initializer(0, shape=shp_bias_cls, dtype=self.ms_type).init_data()
 
         shp_weight_reg = (num_anchors * 4, feat_channels, 1, 1)
         shp_bias_reg = (num_anchors * 4,)
-        weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=self.ms_type).to_tensor()
-        bias_reg = initializer(0, shape=shp_bias_reg, dtype=self.ms_type).to_tensor()
+        weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=self.ms_type).init_data()
+        bias_reg = initializer(0, shape=shp_bias_reg, dtype=self.ms_type).init_data()
 
         for i in range(num_layers):
             rpn_reg_cls_block = RpnRegClsBlock(in_channels, feat_channels, num_anchors, cls_out_channels, \

research/cv/res2net_faster_rcnn/src/FasterRcnn/fpn_neck.py

@@ -34,7 +34,7 @@ def _conv(
     shape = (out_channels, in_channels, kernel_size, kernel_size)
     weights = initializer(
         "XavierUniform", shape=shape, dtype=mstype.float32
-    ).to_tensor()
+    ).init_data()
     shape_bias = (out_channels,)
     biass = bias_init_zeros(shape_bias)
     return nn.Conv2d(

research/cv/res2net_faster_rcnn/src/FasterRcnn/rcnn.py

@@ -101,11 +101,11 @@ class Rcnn(nn.Cell):
         shape_0 = (self.rcnn_fc_out_channels, representation_size)
         weights_0 = initializer(
             "XavierUniform", shape=shape_0[::-1], dtype=self.ms_type
-        ).to_tensor()
+        ).init_data()
         shape_1 = (self.rcnn_fc_out_channels, self.rcnn_fc_out_channels)
         weights_1 = initializer(
             "XavierUniform", shape=shape_1[::-1], dtype=self.ms_type
-        ).to_tensor()
+        ).init_data()
         self.shared_fc_0 = DenseNoTranpose(
             representation_size, self.rcnn_fc_out_channels, weights_0
         )
@@ -117,12 +117,12 @@ class Rcnn(nn.Cell):
             "Normal",
             shape=[num_classes, self.rcnn_fc_out_channels][::-1],
             dtype=self.ms_type,
-        ).to_tensor()
+        ).init_data()
         reg_weight = initializer(
             "Normal",
             shape=[num_classes * 4, self.rcnn_fc_out_channels][::-1],
             dtype=self.ms_type,
-        ).to_tensor()
+        ).init_data()
         self.cls_scores = DenseNoTranpose(
             self.rcnn_fc_out_channels, num_classes, cls_weight
         )

research/cv/res2net_faster_rcnn/src/FasterRcnn/rpn.py

@@ -220,22 +220,22 @@ class RPN(nn.Cell):
         shp_bias_conv = (feat_channels,)
         weight_conv = initializer(
             "Normal", shape=shp_weight_conv, dtype=self.ms_type
-        ).to_tensor()
-        bias_conv = initializer(0, shape=shp_bias_conv, dtype=self.ms_type).to_tensor()
+        ).init_data()
+        bias_conv = initializer(0, shape=shp_bias_conv, dtype=self.ms_type).init_data()
 
         shp_weight_cls = (num_anchors * cls_out_channels, feat_channels, 1, 1)
         shp_bias_cls = (num_anchors * cls_out_channels,)
         weight_cls = initializer(
             "Normal", shape=shp_weight_cls, dtype=self.ms_type
-        ).to_tensor()
-        bias_cls = initializer(0, shape=shp_bias_cls, dtype=self.ms_type).to_tensor()
+        ).init_data()
+        bias_cls = initializer(0, shape=shp_bias_cls, dtype=self.ms_type).init_data()
 
         shp_weight_reg = (num_anchors * 4, feat_channels, 1, 1)
         shp_bias_reg = (num_anchors * 4,)
         weight_reg = initializer(
             "Normal", shape=shp_weight_reg, dtype=self.ms_type
-        ).to_tensor()
-        bias_reg = initializer(0, shape=shp_bias_reg, dtype=self.ms_type).to_tensor()
+        ).init_data()
+        bias_reg = initializer(0, shape=shp_bias_reg, dtype=self.ms_type).init_data()
 
         for i in range(num_layers):
             rpn_reg_cls_block = RpnRegClsBlock(

research/cv/rfcn/src/rfcn/rpn.py

@@ -166,18 +166,18 @@ class RPN(nn.Cell):
 
         shp_weight_conv = (feat_channels, in_channels, 3, 3)
         shp_bias_conv = (feat_channels,)
-        weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=self.ms_type).to_tensor()
-        bias_conv = initializer(0, shape=shp_bias_conv, dtype=self.ms_type).to_tensor()
+        weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=self.ms_type).init_data()
+        bias_conv = initializer(0, shape=shp_bias_conv, dtype=self.ms_type).init_data()
 
         shp_weight_cls = (num_anchors * cls_out_channels, feat_channels, 1, 1)
         shp_bias_cls = (num_anchors * cls_out_channels,)
-        weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=self.ms_type).to_tensor()
-        bias_cls = initializer(0, shape=shp_bias_cls, dtype=self.ms_type).to_tensor()
+        weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=self.ms_type).init_data()
+        bias_cls = initializer(0, shape=shp_bias_cls, dtype=self.ms_type).init_data()
 
         shp_weight_reg = (num_anchors * 4, feat_channels, 1, 1)
         shp_bias_reg = (num_anchors * 4,)
-        weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=self.ms_type).to_tensor()
-        bias_reg = initializer(0, shape=shp_bias_reg, dtype=self.ms_type).to_tensor()
+        weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=self.ms_type).init_data()
+        bias_reg = initializer(0, shape=shp_bias_reg, dtype=self.ms_type).init_data()
 
         for i in range(num_layers):
             rpn_reg_cls_block = RpnRegClsBlock(in_channels, feat_channels, num_anchors, cls_out_channels, \