class Add(Layer):
def __init__(self, op, op_type, tflite_interpreter):
Layer.__init__(self, op, op_type, tflite_interpreter)
self.tflite_add_parser = AddOptions()
self.tflite_add_parser.Init(self.op.BuiltinOptions().Bytes, self.op.BuiltinOptions().Pos)
def generate(self):
node_output_detail = self.tflite_interpreter._get_tensor_details(self.op.Outputs(0))
prev_node_names = self.input_nodes_name.copy()
for input_idx in range(self.op.InputsLength()):
node_input_detail = self.tflite_interpreter._get_tensor_details(self.op.Inputs(input_idx))
if node_input_detail['shape'].size == 4:
# it's a op node, do nothing
continue
prev_node_name = node_input_detail['name']
# it's a parameter node, create constant node
if node_input_detail['shape'].size == 0:
# add all
tensor_node = make_onnx_constant_number(self.tflite_interpreter, node_input_detail)
self.weight_node_list.append(tensor_node)
elif node_input_detail['shape'].size == 1:
# channelwise add
tensor_node = make_onnx_channelwise_constant_number(self.tflite_interpreter, node_input_detail, shape=self.node_output_shape)
self.weight_node_list.append(tensor_node)
if prev_node_name not in prev_node_names:
prev_node_names.append(prev_node_name)
add_node_name = self.node_name
add_node = onnx.helper.make_node(
'Add',
inputs=prev_node_names,
outputs=[add_node_name],
name=add_node_name
)
# update tables
self.node_list.append(add_node)
# change output node's input_name
for o_n in self.output_nodes:
for idx, o_n_i_n in enumerate(o_n.input_nodes_name):
if o_n_i_n == self.node_name:
o_n.input_nodes_name[idx] = self.node_list[-1].name
return self.node_list, self.value_infos, self.weight_node_list, {}
def defuse_activation_function(self):
return defused_activation_node_generator(
activation_function_type=self.tflite_add_parser.FusedActivationFunction(),
op=self.op,
tflite_interpreter=self.tflite_interpreter)
def _convert_elemwise(self, relay_op, op):
"""Generic method to Convert TFLite elemwise"""
try:
from tflite.Operator import Operator
from tflite.AddOptions import AddOptions
from tflite.SubOptions import SubOptions
from tflite.MulOptions import MulOptions
from tflite.DivOptions import DivOptions
from tflite.BuiltinOptions import BuiltinOptions
from tflite.ActivationFunctionType import ActivationFunctionType
except ImportError:
raise ImportError("The tflite package must be installed")
assert isinstance(op, Operator)
input_tensors = self.get_input_tensors(op)
assert len(input_tensors) == 2, "input tensors length should be 2"
lhs_tensor = input_tensors[0]
lhs_expr = self.get_expr(lhs_tensor.tensor_idx)
rhs_tensor = input_tensors[1]
if self.has_expr(rhs_tensor.tensor_idx):
# In most cases, we can assume that TOCO fuses elemwise operators
# with constants - it means both will be tensors.
rhs_expr = self.get_expr(rhs_tensor.tensor_idx)
else:
# However, in some corner cases, the elemwise operator is not fused,
# we can receive as constant.
rhs_type_str = self.get_tensor_type_str(rhs_tensor.tensor.Type())
rhs_expr = self.exp_tab.new_const(
self.get_tensor_value(rhs_tensor), dtype=rhs_type_str)
out = relay_op(lhs_expr, rhs_expr)
# Options (fused_activation_function)
options = None
if op.BuiltinOptionsType() == BuiltinOptions.AddOptions:
options = AddOptions()
elif op.BuiltinOptionsType() == BuiltinOptions.SubOptions:
options = SubOptions()
elif op.BuiltinOptionsType() == BuiltinOptions.MulOptions:
options = MulOptions()
elif op.BuiltinOptionsType() == BuiltinOptions.DivOptions:
options = DivOptions()
if options is not None:
op_options = op.BuiltinOptions()
options.Init(op_options.Bytes, op_options.Pos)
fused_activation_fn = options.FusedActivationFunction()
# if we have activation fn
if fused_activation_fn != ActivationFunctionType.NONE:
out = self.convert_fused_activation_function(
out, fused_activation_fn)
return out
