fix pad converter and support padv2

tools/python/transform/tensorflow_converter.py

@@ -102,6 +102,7 @@ TFSupportedOps = [
     'DepthToSpace',
     'SpaceToDepth',
     'Pad',
+    'PadV2',
     'ConcatV2',
     'Mean',
     'Prod',
@@ -252,6 +253,7 @@ class TensorflowConverter(base_converter.ConverterInterface):
             TFOpType.DepthToSpace.name: self.convert_space_depth,
             TFOpType.SpaceToDepth.name: self.convert_space_depth,
             TFOpType.Pad.name: self.convert_pad,
+            TFOpType.PadV2.name: self.convert_pad,
             TFOpType.ConcatV2.name: self.convert_concat,
             TFOpType.Const.name: self.convert_nop,
             TFOpType.Gather.name: self.convert_gather,
@@ -785,13 +787,19 @@ class TensorflowConverter(base_converter.ConverterInterface):
         pad_type_arg = op.arg.add()
         pad_type_arg.name = MaceKeyword.mace_pad_type_str
 
-        if tf_op.type == TFOpType.Pad:
+        if tf_op.type == TFOpType.Pad or tf_op.type == TFOpType.PadV2:
             if len(tf_op.inputs) == 3:
                 constant_value_arg = op.arg.add()
                 constant_value_arg.name = MaceKeyword.mace_constant_value_str
-                constant_value = tf_op.inputs[2].eval().astype(np.int32) \
-                    .flat[0]
-                constant_value_arg.i = constant_value
+                constant_value = tf_op.inputs[2].eval().flat[0]
+                tf_dt = tf_op.inputs[2].dtype
+                if tf_dt == tf.float32:
+                    constant_value_arg.f = constant_value
+                elif tf_dt == tf.int32:
+                    constant_value_arg.i = constant_value
+                else:
+                    mace_check(False,
+                               "Unsupported data type: %s" % tf_dt.name)
                 self._skip_tensor.add(tf_op.inputs[2].name)
 
             pad_type_arg.i = PadType.CONSTANT.value