_operators.py

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from __future__ import unicode_literals

import numpy as np
import copy

from typing import Sequence, Callable, List, Tuple, Optional, Text, Any
from coremltools.models.neural_network import NeuralNetworkBuilder  #type: ignore
from ._graph import Node, Graph
from coremltools.proto import NeuralNetwork_pb2 #type: ignore
from ._error_utils import ErrorHandling

INT_MAX = 2**30

'''
General common functions
'''

def _compare(a, b, encoding="utf8"): #type: (Text, Text, Text) -> bool
    if isinstance(a, bytes):
        a = a.decode(encoding)
    if isinstance(b, bytes):
        b = b.decode(encoding)
    return a == b

def _is_input_shape_mapping_defined(node, graph): # type: (Node, Graph) -> Bool
    if node.inputs[0] in graph.onnx_coreml_shape_mapping:
        return True
    else:
        return False

def _update_shape_mapping_unchanged(node, graph, err): # type: (Node, Graph, ErrorHandling) -> None
    if _is_input_shape_mapping_defined(node, graph):
        graph.onnx_coreml_shape_mapping[node.outputs[0]] = graph.onnx_coreml_shape_mapping[node.inputs[0]]


def _convert_broadcast_op(builder, node, graph, err, mode): # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling, Text) -> None
    if node.op_type == 'Max' or node.op_type == 'Min' or node.op_type == 'Mean':
        if len(node.inputs) == 1:
            inputs = [node.inputs[0], node.inputs[0]]
        else:
            inputs = node.inputs
    else:
        inputs = node.inputs

    if node.op_type == 'Sub':
        builder.add_elementwise(
            name=node.name + '_neg',
            input_names=[inputs[1]],
            output_name=inputs[1] + '_neg',
            mode='MULTIPLY',
            alpha=-1.0
        )
        builder.add_elementwise(
            name=node.name,
            input_names=[inputs[0], inputs[1] + '_neg'],
            output_name=node.outputs[0],
            mode=mode
        )
    else:
        builder.add_elementwise(
            name=node.name,
            input_names=inputs,
            output_name=node.outputs[0],
            mode=mode
        )

    if _is_input_shape_mapping_defined(node, graph):
        ranks = [len(graph.onnx_coreml_shape_mapping[input_]) for input_ in node.inputs]
        max_id = np.argmax(np.array(ranks))
        graph.onnx_coreml_shape_mapping[node.outputs[0]] = graph.onnx_coreml_shape_mapping[node.inputs[max_id]]


def _get_coreml_target_shape(target_shape, builder, node, graph, err):
    # type: (Tuple[int, ...], NeuralNetworkBuilder, node, Graph, ErrorHandling) -> Optional[Tuple[int, ...]]

    if len(target_shape) == 1:  # (D,)
        coreml_shape = (1, target_shape[0], 1, 1)  # type: Optional[Tuple[int, ...]]
        if _is_input_shape_mapping_defined(node, graph):
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = [2]
    elif len(target_shape) == 2:  # (S,D)
        coreml_shape = target_shape + (1, 1)
        if _is_input_shape_mapping_defined(node, graph):
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = [0,2]
    elif len(target_shape) == 3:  # (C,H,W)
        coreml_shape = (
            1, target_shape[0], target_shape[1], target_shape[2]
        )
        if _is_input_shape_mapping_defined(node, graph):
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = [2,3,4]
    elif len(target_shape) == 4:
        coreml_shape = target_shape
        if _is_input_shape_mapping_defined(node, graph):
            mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
            if mapp[0] == 1 and coreml_shape[0] == 1:
                graph.onnx_coreml_shape_mapping[node.outputs[0]] = [1,2,3,4]
            else:
                graph.onnx_coreml_shape_mapping[node.outputs[0]] = [0,2,3,4]
    elif len(target_shape) > 4:
        #return err.unsupported_op_configuration(builder, node, graph, "Supports tensors not more than 4d")  # type: ignore
        diff = len(target_shape) - 4
        if all([d == 1 for d in target_shape[:diff]]):
            coreml_shape = target_shape[diff:]
        else:
            err.unsupported_op_configuration(builder, node, graph, "Tensors more than rank 4 are not supported")  # type: ignore
        if _is_input_shape_mapping_defined(node, graph):
            if target_shape[0] == 1 and len(target_shape) == 5:
                graph.onnx_coreml_shape_mapping[node.outputs[0]] = [1,0,2,3,4]
        else:
            return err.unsupported_op_configuration(builder, node, graph, "Supports tensors not more than 4d") # type: ignore
    else:
        coreml_shape = None
    return coreml_shape

def _get_coreml_axis(axes, builder, node, graph, err): # type: (List[int], NeuralNetworkBuilder, node, Graph, ErrorHandling) -> Text
    coreml_axis = ""
    if node.inputs[0] not in graph.shape_dict:
        return err.unsupported_op_configuration(builder, node, graph, "Failed to translate axis")
    input_shape = graph.shape_dict[node.inputs[0]]
    if len(input_shape) == 1: coreml_axis = 'C'
    elif len(input_shape) == 2:
        if len(axes) == 1 and axes[0] == 1: coreml_axis = 'C'
    elif len(input_shape) == 3:
        for ind in [['C','H','W'][i] for i in axes]: coreml_axis += ind
    elif len(input_shape) == 4:
        for ind in [['B','C','H','W'][i] for i in axes]: coreml_axis += ind
    return coreml_axis


def _add_transpose_before_after(layer_func, # function for layer conversion
                                input_names, # List[str]
                                output_names, # List[str]
                                transpose_dims, # List[int]
                                **kwargs):  # type: ignore

    for i, input_ in enumerate(input_names):
        kwargs['builder'].add_permute(name=kwargs['node'].name + '_input_transpose' + str(i),
                                      dim=transpose_dims,
                                      input_name=input_,
                                      output_name=kwargs['node'].name + '_' + input_ + '_transpose')

    new_input_names = [kwargs['node'].name + '_' + input_ + '_transpose' for input_ in input_names]
    new_output_names = [output_ + '_transpose' for output_ in output_names]
    layer_func(new_input_names, new_output_names, **kwargs)

    for i, output_ in enumerate(output_names):
        kwargs['builder'].add_permute(name=kwargs['node'].name + '_output_transpose' + str(i),
                                      dim=transpose_dims,
                                      input_name=output_ + '_transpose',
                                      output_name=output_)


def _add_inner_product(input_names, output_names, **kwargs):
    node = kwargs['node']
    builder = kwargs['builder']
    builder.add_inner_product(
        name=node.name,
        W=kwargs['W'],
        b=kwargs['b'],
        input_channels=kwargs['W'].shape[1],
        output_channels=kwargs['W'].shape[0],
        has_bias=kwargs['b'] is not None,
        input_name=input_names[0],
        output_name=output_names[0]
    )

def _add_conv_like_op(add_func, get_params_func, params_dict,
                      builder, node, graph, err):

    if node.inputs[0] in graph.onnx_coreml_shape_mapping:
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]

        r = len(mapp)
        if not (r == 3 or r == 4):
            return err.unsupported_op_configuration(builder, node, graph, "more than 4 axes not supported")
        if r == 4:
            if not (mapp == [1, 2, 3, 4] or mapp == [0, 2, 3, 4]):
                return err.unsupported_op_configuration(builder, node, graph,
                                                        "error in axes alignment between onnx and coreml")
            get_params_func(builder, node, graph, err, params_dict)
            add_func(node.inputs, node.outputs, params_dict=params_dict, node=node, builder=builder, graph=graph, err=err)
        if r == 3:
            if mapp == [1, 2, 3]:  # [B,C,H]
                # spatial dimension: height
                get_params_func(builder, node, graph, err, params_dict, axis='height')
                add_func(node.inputs, node.outputs, params_dict=params_dict, node=node, builder=builder, graph=graph, err=err)
            elif mapp == [1, 2, 4]:  # [B,C,W]
                # spatial dimension: width
                get_params_func(builder, node, graph, err, params_dict, axis='width')
                add_func(node.inputs, node.outputs, params_dict=params_dict, node=node, builder=builder, graph=graph, err=err)
            elif mapp == [2, 3, 4]:  # [C,H,W] in CoreML, but it represents [B,C,D] in ONNX.
                # spatial dimension: sequence
                get_params_func(builder, node, graph, err, params_dict, axis='width')
                node.inputs = [node.inputs[0]]
                _add_transpose_before_after(add_func,
                                            node.inputs,
                                            node.outputs,
                                            [0, 2, 1, 3], # swap C & H
                                            builder=builder, node=node, params_dict=params_dict, graph=graph, err=err)

            elif mapp == [1, 2, 0]:  # [B,C,S]
                # spatial dimension: sequence
                get_params_func(builder, node, graph, err, params_dict, axis='width')
                node.inputs = [node.inputs[0]]
                _add_transpose_before_after(add_func,
                                            node.inputs,
                                            node.outputs,
                                            [3, 1, 2, 0],
                                            builder=builder, node=node, params_dict=params_dict, graph=graph, err=err)
            else:
                return err.unsupported_op_configuration(builder, node, graph,
                                                        "error in axes alignment between onnx and coreml")

    else:
        get_params_func(builder, node, graph, err, params_dict)
        add_func(node.inputs, node.outputs, params_dict=params_dict, builder=builder, node=node, graph=graph, err=err)


def _is_no_op(builder, node, graph, err): # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> Bool

    if node.inputs[0] in graph.shape_dict and node.outputs[0] in graph.shape_dict:
        if graph.shape_dict[node.inputs[0]] == graph.shape_dict[node.outputs[0]]:
            builder.add_activation(
                name=node.name,
                non_linearity='LINEAR',
                input_name=node.inputs[0],
                output_name=node.outputs[0],
                params=[1.0, 0.0]
            )
            _update_shape_mapping_unchanged(node, graph, err)
            return True

    return False

'''
Layer conversion functions
'''

def _convert_abs(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_unary(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        mode='abs'
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_add(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    # check if its equivalent to a bias layer
    if len(node.inputs) > 1:
        if node.inputs[1] in node.input_tensors:
            second_input = np.squeeze(node.input_tensors[node.inputs[1]])
            if len(second_input.shape) == 1:
                builder.add_bias(name=node.name,
                                 b=second_input,
                                 input_name=node.inputs[0],
                                 output_name=node.outputs[0],
                                 shape_bias=[second_input.shape[0]])
                return
    '''
    Supported shapes by CoreML 2.0 for broadcasting (-1 means it can be 1 or greater than 1):
    (i.e. all of the outputs must have one of these shapes for broadcasting support)
    - (S=-1,B=-1,1,1,1)
    - (S=-1,B=-1,C,1,1)
    - (S=-1,B=-1,1,H,W)
    - (S=-1,B=-1,C,H,W)
    Unsupported:
    - (S=-1,B=-1,1,1,W)
    - (S=-1,B=-1,1,H,1)
    - (S=-1,B=-1,C,1,W)
    - (S=-1,B=-1,C,H,1)    
    '''
    _convert_broadcast_op(builder, node, graph, err, "ADD")

def _convert_sub(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    _convert_broadcast_op(builder, node, graph, err, "ADD")


def _get_conv_params(builder, node, graph, err, params_dict, axis=None):
    if 'dilations' not in node.attrs:
        params_dict['dilations'] = [1, 1]
    else:
        if axis == 'height':
            params_dict['dilations'] = node.attrs['dilations']
            params_dict['dilations'].append(1)
        elif axis == 'width':
            params_dict['dilations'] = node.attrs['dilations']
            params_dict['dilations'].insert(0, 1)
        else:
            params_dict['dilations'] = node.attrs['dilations']

    if 'pads' not in node.attrs:
        params_dict['pads'] = [0, 0, 0, 0]
    else:
        pads = node.attrs['pads']
        if axis == 'height':
            pads = [pads[0], 0, pads[1], 0]
        elif axis == 'width':
            pads = [0, pads[0], 0, pads[1]]
        params_dict['pads'] = pads

    if params_dict['is_deconv']:
        params_dict['crops'] = copy.copy(params_dict['pads'])
        params_dict['pads'] = [0, 0, 0, 0]
        if sum(params_dict['crops']) == 0:
            params_dict['is_post_crop'] = False
        else:
            params_dict['is_post_crop'] = True

    if "kernel_shape" in node.attrs:
        params_dict['kernel_shape'] = node.attrs["kernel_shape"]
    else:
        # w_shape is ONNX format shape
        w_shape = params_dict["w_shape"]
        if len(w_shape) == 4:
            params_dict['kernel_shape'] = [w_shape[-2], w_shape[-1]]
        else:
            params_dict['kernel_shape'] = [w_shape[-1]]
    params_dict['strides'] = node.attrs.get("strides", [1, 1] if axis is None else [1])

    if axis == 'height':
        if params_dict['W'] is not None:
            params_dict['W'] = np.expand_dims(params_dict['W'], axis=-1)
        params_dict['kernel_shape'].append(1)
        params_dict['strides'].append(1)
    elif axis == 'width':
        if params_dict['W'] is not None:
            params_dict['W'] = np.expand_dims(params_dict['W'], axis=-2)
        params_dict['strides'].insert(0,1)
        params_dict['kernel_shape'].insert(0,1)

        
    params_dict['out_shape'] = None
    params_dict['padding_type'] = 'valid'
    params_dict['same_padding_asymmetry_mode'] = 'BOTTOM_RIGHT_HEAVY'

    if params_dict['W'] is not None:
        if not params_dict['is_deconv']:
            params_dict['W'] = params_dict['W'].transpose((2, 3, 1, 0))  # type: ignore
        else:
            params_dict['W'] = params_dict['W'].transpose((2, 3, 0, 1))  # type: ignore

    if "auto_pad" in node.attrs and \
            not _compare(node.attrs["auto_pad"], 'VALID'):
        params_dict['padding_type'] = 'same'
        if _compare(node.attrs["auto_pad"], 'SAME_LOWER'):
            params_dict['same_padding_asymmetry_mode'] = 'TOP_LEFT_HEAVY'

    if params_dict['is_deconv']:
        if 'output_shape' in node.attrs:
            if axis == 'height':
                params_dict['out_shape'] = (node.attrs['output_shape'][-1], 1)  # (Hout, wout)
            elif axis == 'width':
                params_dict['out_shape'] = (1, node.attrs['output_shape'][-1])  # (Hout, wout)
            else:
                params_dict['out_shape'] = (node.attrs['output_shape'][-2], node.attrs['output_shape'][-1])  # (Hout, wout)
        elif 'output_padding' in node.attrs:
            post_pads = node.attrs['output_padding']
            if sum(post_pads) != 0:
                t = l = b = r = 0
                if len(post_pads) == 1:
                    if axis == 'height':
                        b = post_pads[0]
                    elif axis == 'width':
                        r = post_pads[0]
                    else:
                        err.unsupported_op_configuration(builder, node, graph,
                                                            "length 1 output padding attribute only supported for 1D conv")
                elif len(post_pads) == 2:
                    if axis == 'height':
                        b, r = post_pads
                    elif axis == 'width':
                        r, b = post_pads
                    else:
                        b, r = post_pads
                elif len(post_pads) == 4:
                    b, r, t, l = post_pads
                else:
                    return err.unsupported_op_configuration(builder, node, graph,
                                                            "Supports only length 1 or 2 or 4 output padding attribute")
                def _update_crop_pad(idx, v):
                    if params_dict['crops'][idx] >= v:
                        params_dict['crops'][idx] -= v
                    else:
                        params_dict['pads'][idx] = v - params_dict['crops'][idx]

                _update_crop_pad(0, t)
                _update_crop_pad(1, l)
                _update_crop_pad(2, b)
                _update_crop_pad(3, r)
                params_dict['is_post_crop'] = True if sum(params_dict['crops']) > 0 else False
                params_dict['is_pre_pad'] = True if sum(params_dict['pads']) > 0 else False

def _add_conv(input_names, output_names, **kwargs):
    params_dict = kwargs['params_dict']
    node = kwargs['node']
    builder = kwargs['builder']
    graph = kwargs['graph']
    err = kwargs['err']

    W_shape = params_dict['w_shape']

    output_name = output_names[0]
    pre_padding_input_name = input_names[0]
    
    if params_dict.get('is_post_crop', False):
        output_name += '_conv_tranpose_post_crop'
    if params_dict.get('is_pre_pad', False):
        input_names[0] += '_conv_tranpose_pre_pad'

    if params_dict['W'] is None and len(node.inputs) == 1:
        return err.unsupported_op_configuration(builder, node, graph, "Kernel weight missing")
    
    if params_dict['is_deconv']:
        oc = W_shape[1] * params_dict['groups']
        kc = W_shape[0]
    else:
        oc = W_shape[0]
        kc = W_shape[1]

    if params_dict.get('is_pre_pad', False):
        builder.add_padding(
            name=node.name + '_pre_pad',  # type: ignore
            left=params_dict['pads'][1],
            right=params_dict['pads'][3],
            top=params_dict['pads'][0],
            bottom=params_dict['pads'][2],
            input_name=pre_padding_input_name,
            output_name=input_names[0],
            value=0
        )
    
    builder.add_convolution(
        name=node.name,
        kernel_channels=kc,
        output_channels=oc,
        height=params_dict['kernel_shape'][0],
        width=params_dict['kernel_shape'][1],
        stride_height=params_dict['strides'][0],
        stride_width=params_dict['strides'][1],
        border_mode=params_dict['padding_type'],
        same_padding_asymmetry_mode=params_dict['same_padding_asymmetry_mode'],
        groups=params_dict['groups'],
        W=params_dict['W'],
        b=params_dict['bias'],
        has_bias=params_dict['bias'] is not None,
        is_deconv=params_dict['is_deconv'],
        output_shape=params_dict['out_shape'],
        input_name=input_names[0] if params_dict['W'] is not None else [input_names[0], input_names[1]],
        output_name=output_name,
        dilation_factors=params_dict['dilations'],
        padding_top=params_dict['pads'][0],
        padding_bottom=params_dict['pads'][2],
        padding_left=params_dict['pads'][1],
        padding_right=params_dict['pads'][3]
    )
    if params_dict.get('is_post_crop', False):
        builder.add_crop(
            name=node.name + '_post_crop',  # type: ignore
            left=params_dict['crops'][1],
            right=params_dict['crops'][3],
            top=params_dict['crops'][0],
            bottom=params_dict['crops'][2],
            input_names=[output_name],
            output_name=output_names[0],
            offset = [0,0]
        )


def _convert_conv(builder, node, graph, err): # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    params_dict = dict()
    #get weights for convolution
    weight_name = node.inputs[1]
    W = None
    if weight_name in node.input_tensors:
        W = node.input_tensors[weight_name]
        params_dict['w_shape'] = W.shape
    else:
        err.missing_initializer(node,
                                "Weight tensor: {} not found in the graph initializer".format(weight_name,))
    params_dict['W'] = W

    params_dict['is_deconv'] = False
    if node.op_type.endswith("Transpose"):
        params_dict['is_deconv'] = True
    bias = None
    if len(node.inputs) > 2:
        bias = node.input_tensors[node.inputs[2]]
    params_dict['bias'] = bias
    params_dict['groups'] = node.attrs.get("group", 1)

    _add_conv_like_op(_add_conv, _get_conv_params, params_dict,
                      builder, node, graph, err)

    # update map
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_relu(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_activation(
        name=node.name,
        non_linearity='RELU',
        input_name=node.inputs[0],
        output_name=node.outputs[0]
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_thresholdedrelu(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    alpha = node.attrs.get('alpha', 1.0)
    builder.add_activation(
        name=node.name,
        non_linearity='THRESHOLDEDRELU',
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        params = alpha
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_reshape(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    shape = tuple(node.attrs.get('shape', ())) # type: (Tuple[int, ...])
    if len(shape) == 0:
        shape_name = node.inputs[1]
        if shape_name in node.input_tensors:
            shape = tuple(node.input_tensors[shape_name].astype(int)) #type: ignore
        else:
            err.missing_initializer(node,
                                "CoreML only supports Reshape layer when the target shape is static and known apriori")

    # check if all entries in shape are 1/-1
    is_flatten = True
    for s in shape:
        if abs(s) != 1:
            is_flatten = False
            break
    if is_flatten:
        builder.add_flatten(
            name=node.name,
            input_name=node.inputs[0],
            output_name=node.outputs[0],
            mode=0
        )
        if _is_input_shape_mapping_defined(node, graph):
            mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
            if len(shape) == 4:
                mapp_out = [mapp[0],2,3,4]
            elif len(shape) == 3:
                mapp_out = [2,3,4]
            elif len(shape) == 2:
                mapp_out = [mapp[0], 2]
            elif len(shape) == 1:
                mapp_out = [2]
            else:
                return err.unsupported_op_configuration(builder, node, graph,
                                                        "Supports only less than equal to 4d tensors")
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = mapp_out
        return

    new_shape = _get_coreml_target_shape(shape, builder, node, graph, err)

    if new_shape is None:
        return err.unsupported_op_configuration(builder, node, graph, "Unsupported shape for reshape")

    builder.add_reshape(
        name=node.name,
        target_shape=new_shape,
        mode=0,
        input_name=node.inputs[0],
        output_name=node.outputs[0]
    )

def _convert_transpose(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    if _is_input_shape_mapping_defined(node, graph):
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        r = len(mapp)
        default_perm = list(range(r))
        default_perm.reverse()
        perm = node.attrs.get("perm", default_perm)
        coreml_perm = []
        for p in perm:
            coreml_perm.append(mapp[p])
        if 1 in mapp:
            batch_index = mapp.index(1)
            batch_index_new = coreml_perm.index(1)
            if batch_index != batch_index_new:
                return err.unsupported_op_configuration(builder, node, graph, "cannot transpose batch dimension")
        perm_translated = []
        for c in coreml_perm:
            if c == 0:
                perm_translated.append(c)
            elif c == 1:
                continue
            else:
                perm_translated.append(c-1)
        perm_final = [-1,-1,-1,-1] # has to be of length 4 corresponding to [S,C,H,W]
        for i in range(4):
            if i not in perm_translated:
                perm_final[i] = i
        if perm_final.count(-1) != len(perm_translated):
            return err.unsupported_op_configuration(builder, node, graph, "unable to translate transpose op to CoreML")
        ctr = 0
        for i, v in enumerate(perm_final):
            if v == -1:
                perm_final[i] = perm_translated[ctr]
                ctr += 1
        perm = tuple(perm_final)
    else:
        perm = node.attrs.get("perm", [0, 3, 2, 1])
        if len(perm) > 4:
            diff = len(perm) - 4
            if all([perm[i] == i for i in range(diff)]):
                perm = [p - diff for p in perm[diff:]]
            else:
                return err.unsupported_op_configuration(builder, node, graph, "Supports only 4d tensors")
        elif len(perm) < 4:
            diff = 4 - len(perm)
            perm = [d for d in range(diff)] + [d + diff for d in perm]
        perm = tuple(perm)

    builder.add_permute(
        name=node.name,
        dim=perm,
        input_name=node.inputs[0],
        output_name=node.outputs[0]
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _get_pool_params(builder, node, graph, err, params_dict, axis=None):
    params_dict['pad_b'], params_dict['pad_l'], params_dict['pad_r'], params_dict['pad_t'] = 0, 0, 0, 0
    params_dict['stride_height'], params_dict['stride_width'] = 1, 1
    params_dict['padding_type'] = 'VALID'
    params_dict['same_padding_asymmetry_mode'] = 'BOTTOM_RIGHT_HEAVY'

    if params_dict['is_global']:
        params_dict['height'], params_dict['width'] = 0, 0
        params_dict['stride_height'], params_dict['stride_width'] = 1, 1
    else:
        kernel_shape = node.attrs["kernel_shape"]
        if axis == 'height':
            params_dict['height'] = kernel_shape[0]
        elif axis == 'width':
            params_dict['width'] = kernel_shape[0]
        else:
            params_dict['height'] = kernel_shape[0]
            params_dict['width'] = kernel_shape[1]

        pads = node.attrs.get('pads', None)
        if pads:
            if axis == 'height':
                params_dict['pad_t'] = pads[0]
                params_dict['pad_b'] = pads[1]
            elif axis == 'width':
                params_dict['pad_l'] = pads[0]
                params_dict['pad_r'] = pads[1]
            else:
                params_dict['pad_t'] = pads[0]
                params_dict['pad_l'] = pads[1]
                params_dict['pad_b'] = pads[2]
                params_dict['pad_r'] = pads[3]

        strides = node.attrs.get('strides', [1, 1])
        if axis == 'height':
            params_dict['stride_height'] = strides[0]
        elif axis == 'width':
            params_dict['stride_width'] = strides[0]
        else:
            params_dict['stride_height'] = strides[0]
            params_dict['stride_width'] = strides[1]

        if "auto_pad" in node.attrs and \
                not _compare(node.attrs["auto_pad"], 'VALID'):
            params_dict['padding_type'] = 'SAME'
            if _compare(node.attrs["auto_pad"], 'SAME_LOWER'):
                params_dict['same_padding_asymmetry_mode'] = 'TOP_LEFT_HEAVY'

    params_dict['exclude_pad_area'] = node.attrs.get('count_include_pad', 0) == 0

def _add_pool(input_names, output_names, **kwargs):
    params_dict = kwargs['params_dict']
    node = kwargs['node']
    kwargs['builder'].add_pooling(
        name=node.name,
        height=params_dict.get('height',1),
        width=params_dict.get('width',1),
        stride_height=params_dict.get('stride_height',1),
        stride_width=params_dict.get('stride_width',1),
        layer_type=params_dict['layer_type'],
        padding_type=params_dict['padding_type'],
        exclude_pad_area=params_dict['exclude_pad_area'],
        is_global=params_dict['is_global'],
        input_name=input_names[0],
        output_name=output_names[0],
        padding_top=params_dict.get('pad_t',0),
        padding_bottom=params_dict.get('pad_b',0),
        padding_left=params_dict.get('pad_l',0),
        padding_right=params_dict.get('pad_r',0),
        same_padding_asymmetry_mode=params_dict['same_padding_asymmetry_mode']
    )


def _convert_pool(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    input_name = node.inputs[0]
    output_name = node.outputs[0]
    params_dict = dict()
    params_dict['is_global'] = False
    if node.op_type.startswith('Global'):
        params_dict['is_global'] = True
    if node.op_type.endswith("MaxPool"):
        params_dict['layer_type'] = "MAX"
    elif node.op_type.endswith("AveragePool"):
        params_dict['layer_type'] = "AVERAGE"
    else:
        return err.unsupported_op_configuration(builder, node, graph, "Unsupported pool type")

    if len(node.outputs) == 2:
        return err.unsupported_op_configuration(builder, node, graph, "argmax with pool unsupported")

    if 'ceil_mode' in node.attrs and node.attrs['ceil_mode'] == 1:
        return err.unsupported_op_configuration(builder, node, graph, "ceil_mod=1 not supported")
    
    if 'dilations' in node.attrs:
        return err.unsupported_op_configuration(builder, node, graph, "dilations not supported")

    _add_conv_like_op(_add_pool, _get_pool_params, params_dict,
                      builder, node, graph, err)

    # update map
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_bn(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    def add_bn(input_names, output_names, **kwargs):
            kwargs['builder'].add_batchnorm(
            name=node.name,
            input_name=input_names[0],
            output_name=output_names[0],
            channels=kwargs['channels'][0],
            gamma=kwargs['scale'],
            beta=kwargs['bias'],
            mean=kwargs['mean'],
            variance=kwargs['var'],
            epsilon=kwargs['epsilon'],)

    if len(node.outputs) > 1:
        return err.unsupported_op_configuration(builder, node, graph, "This converter only supports BatchNormalization with one output")

    epsilon = node.attrs.get("epsilon", 1e-5)
    channels = set()
    for v in node.input_tensors.values():
        channels.add(v.shape)
    assert len(channels) == 1
    channels = channels.pop()
    scale = node.input_tensors[node.inputs[1]] if node.inputs[1] in node.input_tensors else \
            np.ones(shape=channels, dtype=np.float32)
    bias = node.input_tensors[node.inputs[2]] if node.inputs[2] in node.input_tensors else \
            np.zeros(shape=channels, dtype=np.float32)
    mean = node.input_tensors[node.inputs[3]] if node.inputs[3] in node.input_tensors else \
            np.zeros(shape=channels, dtype=np.float32)
    var = node.input_tensors[node.inputs[4]] if node.inputs[4] in node.input_tensors else \
            np.ones(shape=channels, dtype=np.float32)

    mapp = graph.onnx_coreml_shape_mapping.get(node.inputs[0], None)
    if mapp == [2,3,4]:
        _add_transpose_before_after(add_bn,
                                [node.inputs[0]],
                                node.outputs,
                                [0, 2, 1, 3],
                                builder=builder, node=node,scale=scale,bias=bias,mean=mean,var=var,epsilon=epsilon,channels=channels)
    else:
        builder.add_batchnorm(
            name=node.name,
            channels=channels[0],
            gamma=scale,
            beta=bias,
            mean=mean,
            variance=var,
            input_name=node.inputs[0],
            output_name=node.outputs[0],
            epsilon=epsilon
        )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_instancenorm(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    epsilon = node.attrs.get("epsilon", 1e-5)
    scale = node.input_tensors[node.inputs[1]]
    bias = node.input_tensors[node.inputs[2]]

    builder.add_batchnorm(
        name=node.name,
        channels=scale.shape[0],
        gamma=scale,
        beta=bias,
        compute_mean_var=True,
        instance_normalization=True,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        epsilon=epsilon
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_mul(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    _convert_broadcast_op(builder, node, graph, err, "MULTIPLY")

def _convert_mean(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    _convert_broadcast_op(builder, node, graph, err, "AVE")

def _convert_div(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_unary(name=node.name + '_inverse', #type: ignore
                      input_name=node.inputs[1],
                      output_name=node.inputs[1] + '_inverse',
                      mode='inverse')
    builder.add_elementwise(
        name=node.name,
        input_names=[node.inputs[0], node.inputs[1] + '_inverse'],
        output_name=node.outputs[0],
        mode="MULTIPLY"
    )
    if _is_input_shape_mapping_defined(node, graph):
        ranks = [len(graph.onnx_coreml_shape_mapping[input_]) for input_ in node.inputs]
        max_id = np.argmax(np.array(ranks))
        graph.onnx_coreml_shape_mapping[node.outputs[0]] = graph.onnx_coreml_shape_mapping[node.inputs[max_id]]

def _convert_leaky_relu(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    alpha = node.attrs.get('alpha', 0.01)
    builder.add_activation(
        name=node.name,
        non_linearity='LEAKYRELU',
        params=[alpha],
        input_name=node.inputs[0],
        output_name=node.outputs[0]
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_concat(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    def _add_concat(input_names, output_names, **kwargs):
        kwargs['builder'].add_elementwise(
            name=kwargs['node'].name,
            input_names=input_names,
            output_name=output_names[0],
            mode=kwargs['mode']
        )

    axis = node.attrs.get("axis", 1)
    parent_op_type = graph.blob_from_op_type.get(node.inputs[0], None)

    if _is_input_shape_mapping_defined(node, graph):
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        caxis = mapp[axis]
        if caxis == 0:
            _add_concat(node.inputs, node.outputs, node=node, builder=builder, mode='SEQUENCE_CONCAT')
        elif caxis == 2:
            _add_concat(node.inputs, node.outputs, node=node, builder=builder, mode='CONCAT')
        elif caxis == 3:
            _add_transpose_before_after(_add_concat, node.inputs, node.outputs, [0,2,1,3], mode='CONCAT', node=node, builder=builder,)
        elif caxis == 4:
            _add_transpose_before_after(_add_concat, node.inputs, node.outputs, [0,3,2,1], mode='CONCAT', node=node, builder=builder,)
        else:
            return err.unsupported_op_configuration(builder, node, graph, "Concat not supported along batch axis")
    else:
        mode = None
        first_input_shape = None
        if node.inputs[0] in graph.shape_dict:
            first_input_shape = graph.shape_dict[node.inputs[0]]
            if parent_op_type in _SEQUENCE_LAYERS_REGISTRY and \
                len(first_input_shape) == 3:
                if axis == 0:
                    mode = 'SEQUENCE_CONCAT'
                if axis == 2:
                    mode = 'CONCAT'
            elif (len(first_input_shape) == 1 and axis == 0) or \
                (len(first_input_shape) == 3 and axis == 0) or \
                (len(first_input_shape) == 4 and axis == 1) or \
                (len(first_input_shape) == 2 and axis == 1):
                mode = 'CONCAT'
        else: # shape info is not available. Fall back to guessing (ideally this should not happen)
            if axis == 0:
                mode = "SEQUENCE_CONCAT"
            elif axis == 1:
                mode = "CONCAT"
        if mode is None:
            return err.unsupported_op_configuration(builder, node, graph,
                                                    "Unsupported axis {} in input of shape".
                                                    format(axis, str(first_input_shape)))
        _add_concat(node.inputs, node.outputs, node=node, builder=builder, mode=mode)

    _update_shape_mapping_unchanged(node, graph, err)

def _convert_split(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    def _add_split(input_names, output_names, **kwargs):
        kwargs['builder'].add_split(
            name=kwargs['node'].name,
            input_name=input_names[0],
            output_names=output_names
        )

    axis = node.attrs.get("axis", 0)
    splits = node.attrs.get('split', None)
    # check that splits are equal
    if splits:
        if splits.count(splits[0]) != len(splits):
            return err.unsupported_op_configuration(builder, node, graph, 'Only Equal splits are supported')

    if _is_input_shape_mapping_defined(node, graph):
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        if mapp[axis] == 2:
            _add_split(node.inputs, node.outputs, node=node, builder=builder)
        elif mapp[axis] == 0:
            _add_transpose_before_after(_add_split, node.inputs, node.outputs, [1,0,2,3], builder=builder, node=node)
        elif mapp[axis] == 3:
            _add_transpose_before_after(_add_split, node.inputs, node.outputs, [0,2,1,3], builder=builder, node=node)
        elif mapp[axis] == 4:
            _add_transpose_before_after(_add_split, node.inputs, node.outputs, [0,3,2,1], builder=builder, node=node)
        else:
            err.unsupported_op_configuration(builder, node, graph, 'Split along Batch axis not supported')
    else:
        if not (axis == 0 or axis == 1):
            return err.unsupported_op_configuration(builder, node, graph, "Unsupported axis {}".format(axis, ))
        _add_split(node.inputs, node.outputs, node=node, builder=builder)

    if _is_input_shape_mapping_defined(node, graph):
        for out_ in node.outputs:
            graph.onnx_coreml_shape_mapping[out_] = graph.onnx_coreml_shape_mapping[node.inputs[0]]

def _convert_argmax(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    def _add_argmax_or_argmin(input_names, output_names, **kwargs):
        input_name = input_names[0]
        output_name = output_names[0]
        if kwargs['node'].op_type == 'ArgMin':
            kwargs['builder'].add_elementwise(name=kwargs['node'].name + '_multiply_minus_1',  # type: ignore
                                    input_names=[input_name],
                                    output_name=input_name + '_multiply_minus_1',
                                    mode='MULTIPLY',
                                    alpha=-1)
            input_name += '_multiply_minus_1'
        kwargs['builder'].add_reduce(name=kwargs['node'].name,
                           input_name=input_name, output_name=output_name,
                           axis=kwargs['coreml_axis'],
                           mode='argmax')
    '''
    Conversion
    '''
    axis = node.attrs.get('axis', 0)
    keepdims = node.attrs.get('keepdims', 1)

    input_name = node.inputs[0]
    output_name = node.outputs[0]

    if _is_input_shape_mapping_defined(node, graph):
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        coreml_axis = mapp[axis]
        coreml_axis_string = "C"
        if coreml_axis == 1: # coreml_axis corresponds to the batch dimension
            return err.unsupported_op_configuration(builder, node, graph,"Cannot apply operation along Batch axis")
        if coreml_axis != 0:
            coreml_axis_string = ['C','H','W'][coreml_axis-2]
            _add_argmax_or_argmin([input_name], [output_name], builder=builder,node=node,coreml_axis=coreml_axis_string)
        else: # coreml_axis corresponds to the sequence dimension
            _add_transpose_before_after(_add_argmax_or_argmin,
                                        [input_name],
                                        [output_name],
                                        [1,0,2,3],
                                        builder=builder,node=node,coreml_axis=coreml_axis_string)

    else:
        coreml_axis_string = _get_coreml_axis([axis], builder, node, graph, err)
        if coreml_axis_string not in ['C', 'H', 'W', 'HW', 'CHW']:
            return err.unsupported_op_configuration(builder, node, graph,
                                                    "Unable to translate axes attribute to CoreML axis parameter for %s" % axis)
        _add_argmax_or_argmin([input_name], [output_name], builder=builder, node=node, coreml_axis=coreml_axis_string)

    '''
    update output shape map
    '''
    if _is_input_shape_mapping_defined(node, graph):
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        if keepdims == 1:
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = mapp
        else:
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = mapp[:axis] + mapp[axis+1:]

def _convert_reduce(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    # CoreML reduction supported along: C, H, W, CHW, HW

    def _add_reduce(input_names, output_names, **kwargs):
        input_name = input_names[0]
        output_name = output_names[0]

        if 'add_log' in kwargs and kwargs['add_log']:
            if kwargs['node'].op_type == 'ReduceLogSum':
                output_name = output_names[0] + '_before_log'

        kwargs['builder'].add_reduce(name=kwargs['node'].name + '_' + output_name,
                                     input_name=input_name, output_name=output_name,
                                     axis=kwargs['coreml_axis'],
                                     mode=kwargs['mode'])

        if 'add_log' in kwargs and kwargs['add_log']:
            if node.op_type == 'ReduceLogSum':
                kwargs['builder'].add_unary(name=kwargs['node'].name + '_log',
                                        input_name=output_name,
                                        output_name=output_names[0],
                                        mode='log')

    '''
    Conversion
    '''
    input_name = node.inputs[0]
    output_name = node.outputs[0]

    axes = node.attrs.get('axes', None)
    keepdims = node.attrs.get('keepdims', 1)

    if axes is None:
        if node.inputs[0] in graph.onnx_coreml_shape_mapping:
            axes = range(0, len(graph.onnx_coreml_shape_mapping[node.inputs[0]]))
        elif node.inputs[0] in graph.shape_dict:
            axes = range(0, len(graph.shape_dict[node.inputs[0]]))
        else:
            return err.unsupported_op_configuration(builder, node, graph, "Shape inference failed for reduce op")

    if node.op_type == 'ReduceMean':
        mode = 'avg'
    elif node.op_type == 'ReduceL1':
        mode = 'L1'
    elif node.op_type == 'ReduceL2':
        mode = 'L2'
    elif node.op_type == 'ReduceLogSum':
        mode = 'sum'
    elif node.op_type == 'ReduceMax':
        mode = 'max'
    elif node.op_type == 'ReduceMin':
        mode = 'min'
    elif node.op_type == 'ReduceProd':
        mode = 'prod'
    elif node.op_type == 'ReduceSum':
        mode = 'sum'
    elif node.op_type == 'ReduceSumSquare':
        mode = 'sumsquare'
    else:
        return err.unsupported_op_configuration(builder, node, graph, "Unsupported op")

    if _is_input_shape_mapping_defined(node, graph):
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        coreml_axis = ""
        for ind in [['S','B','C','H','W'][mapp[i]] for i in axes]: coreml_axis += ind
        coreml_axis = ''.join(sorted(coreml_axis))
    else:
        coreml_axis = _get_coreml_axis(axes, builder, node, graph, err)

    if coreml_axis in ['C', 'H', 'W', 'HW', 'CHW']:
        _add_reduce([input_name], [output_name], builder=builder, node=node, coreml_axis=coreml_axis, mode=mode, add_log=True)
    else:
        if node.op_type in ['ReduceMean']:
            return err.unsupported_op_configuration(builder, node, graph,
                                                    "Unable to translate axes attribute to CoreML axis parameter for %s" % axes)
        n = len(coreml_axis)
        for i, ax in enumerate(coreml_axis):
            if ax not in ['C', 'H', 'W']:
                return err.unsupported_op_configuration(builder, node, graph,
                                                    "Unable to translate axes attribute to CoreML axis parameter for %s" % axes)
            else:
                if i == 0:
                    iname = input_name
                else:
                    iname = input_name + str(i)
                if i == n-1:
                    oname = output_name
                else:
                    oname = input_name + str(i+1)
                if i < n-1:
                    _add_reduce([iname], [oname], builder=builder, node=node, coreml_axis=ax, mode=mode, add_log=False)
                else:
                    _add_reduce([iname], [oname], builder=builder, node=node, coreml_axis=ax, mode=mode, add_log=True)

    '''
    update output shape map
    '''
    if _is_input_shape_mapping_defined(node, graph):
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        if keepdims == 1:
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = mapp
        else:
            out_mapp = []
            for i, m in enumerate(mapp):
                if i not in axes:
                    out_mapp.append(m)
            if len(out_mapp) == 0:
                out_mapp = [2]
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = out_mapp

def _convert_softmax(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    def _add_softmax(input_names, output_names, **kwargs):
        node = kwargs['node']
        builder = kwargs['builder']

        if node.op_type == 'LogSoftmax':
            builder.add_softmax(
                name=node.name + '_softmax',  # type: ignore
                input_name=node.inputs[0],
                output_name=node.outputs[0] + '_softmax'
            )
            builder.add_unary(
                name=node.name,
                input_name=node.outputs[0] + '_softmax',
                output_name=node.outputs[0],
                mode='log'
            )
        else:
            builder.add_softmax(
                name=node.name,
                input_name=input_names[0],
                output_name=output_names[0]
            )

    axis = node.attrs.get('axis', 1)
    if axis != 1:
        return err.unsupported_op_configuration(builder, node, graph, "Unsupported axis {} for softmax".format(axis, ))

    _add_softmax(node.inputs, node.outputs, node=node, builder=builder)

    if node.inputs[0] in graph.onnx_coreml_shape_mapping:
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        graph.onnx_coreml_shape_mapping[node.outputs[0]] = mapp

def _convert_gemm(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    '''
    operation:  alpha * (A * B) + beta * C
    so far the case only handled is :
    - B is a constant matrix
    - C is a constant vector
    - alpha == beta == 1.0
    - transA is off
    '''

    if node.attrs.get("transA",0) != 0:
        return err.unsupported_op_configuration(builder, node, graph, "This Gemm layer cannot be converted to CoreML inner_product layer")

    if abs(node.attrs.get('alpha',1.0)-1.0) > 1e-3 or abs(node.attrs.get('beta',1.0)-1.0) > 1e-3:
        return err.unsupported_op_configuration(builder, node, graph, "This Gemm layer cannot be converted to CoreML inner_product layer")

    weight_name = node.inputs[1]
    if weight_name in node.input_tensors:
        W = node.input_tensors[weight_name]
        if not node.attrs.get("transB",0):
            W = np.transpose(W)
    else:
        err.missing_initializer(node, "Second input to Gemm layer must be a constant")

    b = None
    if len(node.inputs) > 2:
        b = (node.input_tensors[node.inputs[2]]).flatten()
    if len(W.shape) != 2 or (b is not None and len(b.shape) != 1):
        return err.unsupported_op_configuration(builder, node, graph, "This Gemm layer cannot be converted to CoreML inner_product layer")

    if b is not None:
        if W.shape[0] != b.shape[0]:
            return err.unsupported_op_configuration(builder, node, graph, "This Gemm layer cannot be converted to CoreML inner_product layer")

    if node.inputs[0] in graph.onnx_coreml_shape_mapping:
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        if mapp == [1,2] or mapp == [0,2]: #[B,C] or [S,C]
            _add_inner_product([node.inputs[0]], node.outputs, W=W, b=b, node=node, builder=builder)
        elif mapp == [3,4]: #[H,W]
            _add_transpose_before_after(_add_inner_product, [node.inputs[0]], node.outputs, [2,3,0,1],W=W, b=b, node=node, builder=builder)
        elif mapp == [2,3]: #(C,H)
            _add_transpose_before_after(_add_inner_product, [node.inputs[0]], node.outputs, [1,2,0,3],W=W, b=b, node=node, builder=builder)
        elif mapp == [2,4]: #(C,W)
            _add_transpose_before_after(_add_inner_product, [node.inputs[0]], node.outputs, [1,3,2,0], W=W, b=b, node=node, builder=builder)
        else:
            return err.unsupported_op_configuration(builder, node, graph, "CoreML incompatible axis placement")
    else:
        _add_inner_product([node.inputs[0]], node.outputs , W=W, b=b, node=node, builder=builder)

    if node.inputs[0] in graph.onnx_coreml_shape_mapping:
        graph.onnx_coreml_shape_mapping[node.outputs[0]] = graph.onnx_coreml_shape_mapping[node.inputs[0]]

def _convert_matmul(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    weight_name = node.inputs[1]
    if weight_name in node.input_tensors:
        W = node.input_tensors[weight_name]
    else:
        err.missing_initializer(node, "Second input to Matmul layer must be a constant")

    if len(W.shape) != 2:
        return err.unsupported_op_configuration(builder, node, graph, "This Matmul layer cannot be converted to CoreML inner_product layer")

    W = np.transpose(W)

    if node.inputs[0] in graph.onnx_coreml_shape_mapping:
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        if mapp == [1,2] or mapp == [0,2]: #[B,C] or [S,C]
            _add_inner_product([node.inputs[0]], node.outputs, W=W, b=None, node=node, builder=builder)
        elif mapp == [3,4]: #[H,W]
            _add_transpose_before_after(_add_inner_product, [node.inputs[0]], node.outputs, [2,3,0,1],W=W, b=None, node=node, builder=builder)
        elif mapp == [2,3]: #(C,H)
            _add_transpose_before_after(_add_inner_product, [node.inputs[0]], node.outputs, [1,2,0,3],W=W, b=None, node=node, builder=builder)
        elif mapp == [2,4]: #(C,W)
            _add_transpose_before_after(_add_inner_product, [node.inputs[0]], node.outputs, [1,3,2,0], W=W, b=None, node=node, builder=builder)
        else:
            return err.unsupported_op_configuration(builder, node, graph, "CoreML incompatible axis placement")
    else:
        _add_inner_product([node.inputs[0]], node.outputs , W=W, b=None, node=node, builder=builder)

    if node.inputs[0] in graph.onnx_coreml_shape_mapping:
        graph.onnx_coreml_shape_mapping[node.outputs[0]] = graph.onnx_coreml_shape_mapping[node.inputs[0]]

def _convert_lrn(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    alpha = node.attrs.get("alpha", 1.0e-4)
    beta = node.attrs.get("beta", 0.75)
    bias = node.attrs.get("bias", 1.0)
    size = node.attrs["size"]
    builder.add_lrn(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        alpha=alpha,
        beta=beta,
        k=bias,
        local_size=size
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_sigmoid(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_activation(
        name=node.name,
        non_linearity='SIGMOID',
        input_name=node.inputs[0],
        output_name=node.outputs[0]
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_sign(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_activation(
        name=node.name,
        non_linearity='SIGMOID_HARD',
        input_name=node.inputs[0],
        output_name=node.outputs[0] + '_step',
        params=[10000, 0],
    )
    builder.add_elementwise(name=node.name + '_subtract_half',
                            input_names = node.outputs[0] + '_step',
                            output_name = node.outputs[0] + '_step_half',
                            mode='ADD', alpha=-0.5)
    builder.add_elementwise(name=node.name + '_multiply_2',
                            input_names = node.outputs[0] + '_step_half',
                            output_name = node.outputs[0],
                            mode='MULTIPLY', alpha=2)
    _update_shape_mapping_unchanged(node, graph, err)


def _convert_elu(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    alpha = node.attrs.get('alpha', 1.0)
    builder.add_activation(
        name=node.name,
        non_linearity='ELU',
        params=alpha,
        input_name=node.inputs[0],
        output_name=node.outputs[0]
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_selu(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    alpha = node.attrs.get('alpha', 1.6732)
    gamma = node.attrs.get('gamma', 1.0507)
    builder.add_activation(
        name=node.name + '_elu', #type: ignore
        non_linearity='ELU',
        params=alpha,
        input_name=node.inputs[0],
        output_name=node.inputs[0] + '_elu'
    )
    builder.add_elementwise(
        name=node.name,
        input_names=node.inputs[0] + '_elu',
        output_name=node.outputs[0],
        mode='MULTIPLY',
        alpha=gamma
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_prelu(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    if node.inputs[1] not in node.input_tensors:
        return err.unsupported_op_configuration(builder, node, graph, "Slope must be known!")

    slope = node.input_tensors[node.inputs[1]]
    builder.add_activation(
        name=node.name,
        non_linearity='PRELU',
        params=slope,
        input_name=node.inputs[0],
        output_name=node.outputs[0]
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_tanh(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_activation(
        name=node.name,
        non_linearity='TANH',
        input_name=node.inputs[0],
        output_name=node.outputs[0]
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_pad(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    def _get_pad_params(builder, node, graph, err, params_dict, axis=None):

        pads = node.attrs['pads']
        if not (len(pads) % 2 == 0 and len(pads) >= 2):
            return err.unsupported_op_configuration(builder, node, graph,
                                                    "pads attribute: {}."
                                                    "Length of pads must be a multiple of 2".format(str(pads)))
        if len(pads) == 8:
            az = pads[:2] + pads[4:6]
            if az.count(0) != len(az):
                return err.unsupported_op_configuration(builder, node, graph, "Paddings value {} not supported".format(pads, ))
            pads = pads[2:4] + pads[6:8]

        if len(pads) == 6:
            az = pads[:2] + pads[3:5]
            if az.count(0) != len(az):
                return err.unsupported_op_configuration(builder, node, graph, "Paddings value {} not supported".format(pads, ))
            pads = [pads[2],pads[5]]

        pad_t, pad_b, pad_l, pad_r = 0, 0, 0, 0
        if axis == 'height':
            pad_t, pad_b = pads
        elif axis == 'width':
            pad_l, pad_r = pads
        else:
            pad_t, pad_l, pad_b, pad_r = pads
        params_dict['pad_t'] = pad_t
        params_dict['pad_b'] = pad_b
        params_dict['pad_l'] = pad_l
        params_dict['pad_r'] = pad_r

    def _add_pad(input_names, output_names, **kwargs):
        params_dict = kwargs['params_dict']
        node = kwargs['node']
        builder = kwargs['builder']
        builder.add_padding(
            name=node.name,
            left=params_dict['pad_l'],
            right=params_dict['pad_r'],
            top=params_dict['pad_t'],
            bottom=params_dict['pad_b'],
            value=params_dict['value'],
            input_name=input_names[0],
            output_name=output_names[0],
            padding_type=params_dict['mode']
        )

    params_dict = dict()
    mode = node.attrs['mode']
    if mode == 'reflect' or mode == b'reflect':
        mode = 'reflection'
    elif mode == 'edge' or mode == b'edge':
        mode = 'replication'
    else:
        mode = 'constant'
    params_dict['mode'] = mode
    params_dict['value'] = node.attrs.get('value', 0.0)

    _add_conv_like_op(_add_pad, _get_pad_params, params_dict,
                      builder, node, graph, err)

    # update map
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_slice(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None


    if _is_no_op(builder, node, graph, err):
        return


    def _add_slice(input_names, output_names, **kwargs):
        node = kwargs['node']
        builder = kwargs['builder']
        params_dict = kwargs['params_dict']
        builder.add_slice(
            name=node.name + '_' + output_names[0],
            input_name=input_names[0],
            output_name=output_names[0],
            axis=params_dict['axis'],
            start_index=params_dict['start_index'],
            end_index=params_dict['end_index'],
            stride=1
        )

    params_dict = dict()
    starts = node.attrs['starts']
    ends = node.attrs['ends']
    axes = node.attrs.get('axes', range(len(starts)))

    if node.inputs[0] in graph.shape_dict:
        for ii, _ in enumerate(axes):
            if ends[ii] > INT_MAX:
                ends[ii] = graph.shape_dict[node.inputs[0]][ii]


    if _is_input_shape_mapping_defined(node, graph):
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        r = len(starts)
        for i, ax in enumerate(axes):
            params_dict['start_index'] = starts[i]
            params_dict['end_index'] = ends[i]
            if i == 0:
                iname = node.inputs[0]
            else:
                iname = node.inputs[0] + str(i)
            oname = node.inputs[0] + str(i+1)
            if i == r-1:
                oname = node.outputs[0]

            if mapp[ax] == 2:
                params_dict['axis'] = 'channel'
                _add_slice([iname], [oname], node=node, builder=builder, params_dict=params_dict)
            elif mapp[ax] == 3:
                params_dict['axis'] = 'height'
                _add_slice([iname], [oname], node=node, builder=builder, params_dict=params_dict)
            elif mapp[ax] == 4:
                params_dict['axis'] = 'width'
                _add_slice([iname], [oname], node=node, builder=builder, params_dict=params_dict)
            elif mapp[ax] == 0:
                params_dict['axis'] = 'channel'
                _add_transpose_before_after(_add_slice, [iname], [oname], [1,0,2,3],
                                            node=node, builder=builder, params_dict=params_dict)
            else:
                err.unsupported_op_configuration(builder, node, graph, 'cannot slice along batch axis')
    else:
        params_dict['start_index'] = starts[0]
        params_dict['end_index'] = ends[0]
        input_shape = graph.shape_dict.get(node.inputs[0], None)
        if len(axes) != 1:
            return err.unsupported_op_configuration(builder, node, graph, "Only single axis Slice is supported now")
        if input_shape and len(input_shape) == 4 and len(axes) == 1:
            axis = ['B','channel','height','width'][axes[0]]
        elif len(axes) == 1:
            if axes[0] == 0:
                axis = 'channel'
            elif axes[0] == 1:
                axis = 'height'
            elif axes[0] == 2:
                axis = 'width'
            else:
                return err.unsupported_op_configuration(builder, node, graph, "Slice is supported only along H, W or C dimensions")
        else:
            return err.unsupported_op_configuration(builder, node, graph, "Slice is supported only along one axis for 3D or 4D Tensors")
        params_dict['axis'] = axis
        _add_slice(node.inputs, node.outputs, builder=builder, node=node, params_dict=params_dict)

    _update_shape_mapping_unchanged(node, graph, err)

def _convert_exp(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_unary(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        mode='exp'
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_pow(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    input2 = node.inputs[1]
    is_supported = False
    if input2 in node.input_tensors:
        alpha = node.input_tensors[input2]
        if len(alpha.shape) == 0:
            is_supported = True

    if not is_supported:
        err.missing_initializer(node, "Only mode supported is when the second input is a scalar constant")

    builder.add_unary(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        mode='power',
        alpha = float(alpha)
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_flatten(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    def _add_flatten(input_names, output_names, **kwargs):
        kwargs['builder'].add_flatten(
            name=kwargs['node'].name,
            input_name=input_names[0],
            output_name=output_names[0],
            mode=0
        )

    axis = node.attrs.get('axis', 1)
    if node.inputs[0] in graph.onnx_coreml_shape_mapping:
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        if (mapp[0] == 0 or mapp[0] == 1) and (axis == 0 or axis == 1):
            _add_flatten(node.inputs, node.outputs, builder=builder, node=node)
        elif mapp[0:2] == [0,1] and axis == 2:
            _add_flatten(node.inputs, node.outputs, builder=builder, node=node)
        elif len(mapp) == 1 and axis == 1 and mapp[0] < 4:
            _add_flatten(node.inputs, node.outputs, builder=builder, node=node)
        else:
            return err.unsupported_op_configuration(builder, node, graph, 'Flatten axis mode not supported')
    else:
        _add_flatten(node.inputs, node.outputs, builder=builder, node=node)

    if node.inputs[0] in graph.onnx_coreml_shape_mapping:
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        if len(mapp) == 1 and axis == 1 and mapp[0] < 4:
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = [mapp[0], mapp[0]+1]
        else:
            graph.onnx_coreml_shape_mapping[node.outputs[0]] = [mapp[0],2]

def _convert_max(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    _convert_broadcast_op(builder, node, graph, err, "MAX")

def _convert_min(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    _convert_broadcast_op(builder, node, graph, err, "MIN")

def _convert_softsign(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_activation(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        non_linearity='SOFTSIGN'
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_softplus(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_activation(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        non_linearity='SOFTPLUS'
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_hardsigmoid(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    alpha = node.attrs.get('alpha', 0.2)
    beta = node.attrs.get('beta', 0.5)
    builder.add_activation(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        non_linearity='SIGMOID_HARD',
        params = [alpha, beta]
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_neg(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_elementwise(
        name=node.name,
        input_names=node.inputs,
        output_name=node.outputs[0],
        mode='MULTIPLY',
        alpha=-1.0
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_log(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_unary(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        mode='log'
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_sqrt(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_unary(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        mode='sqrt'
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_reciprocal(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_unary(
        name=node.name,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        mode='inverse'
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_reorganize_data(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    mode = 'SPACE_TO_DEPTH'
    if node.op_type == 'DepthToSpace':
        mode = 'DEPTH_TO_SPACE'
    block_size = node.attrs.get('blocksize', 2)
    builder.add_reorganize_data(name = node.name,
         input_name = node.inputs[0],
         output_name = node.outputs[0],
         mode=mode,
         block_size=block_size
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_upsample(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    if 'scales' in node.attrs:
        scales = node.attrs["scales"]
        if len(scales) != 4 or scales[0] != 1.0 or scales[1] != 1.0:
            err.unsupported_op_configuration(builder, node, graph, "Unsupported scales {} for upsample".format(scales))
        height_scale = int(scales[2])
        width_scale = int(scales[3])
    elif len(node.input_tensors):
        key = next(iter(node.input_tensors.keys()))
        scales = node.input_tensors[key]
        height_scale = int(scales[2])
        width_scale = int(scales[3])
    else:
        if len(node.inputs) > 1:
            return err.unsupported_op_configuration(builder, node, graph,
                                                    "This ONNX upsample layer has 'scales' provided as an input. CoreML upsample requires 'scales' as an attribute of the layer.")
        height_scale = int(node.attrs.get('height_scale', 1))
        width_scale = int(node.attrs.get('width_scale', 1))
    mode_convert = {
        "nearest": "NN",
        "linear": "BILINEAR",
    }
    mode = mode_convert[node.attrs["mode"].decode("UTF-8")]
    builder.add_upsample(
        name=node.name,
        scaling_factor_h=height_scale,
        scaling_factor_w=width_scale,
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        mode=mode,
    )
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_clip(builder, node, graph, err): # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    # clip(x, a, b) = max(min(x, a), b) = -min(-min(x, a), -b)

    if node.attrs.get('max') is None:
        min_limit = node.attrs.get('min', float(-2**16-1))
        builder.add_unary(name=node.name,
                          input_name=node.inputs[0],
                          output_name=node.outputs[0],
                          mode='threshold',
                          alpha=min_limit,
                          shift=0,
                          scale=1.0)
    elif node.attrs.get('min') is None:
        max_limit = node.attrs.get('max', float(2**16-1))
        builder.add_unary(name=node.name + '_min_minus_x_minus_b',
                          input_name=node.inputs[0],
                          output_name=node.inputs[0] + '_min_minus_x_minus_b',
                          mode='threshold',
                          alpha=-max_limit,
                          shift=0,
                          scale=-1.0)

        builder.add_activation(name=node.name,
                               non_linearity='LINEAR',
                               input_name=node.inputs[0] + '_min_minus_x_minus_b',
                               output_name=node.outputs[0],
                               params=[-1.0, 0])

    else:
        min_limit = node.attrs.get('min')
        max_limit = node.attrs.get('max')
        builder.add_unary(name=node.name + '_min_x_a',
                          input_name=node.inputs[0],
                          output_name=node.inputs[0] + '_min_x_a',
                          mode='threshold',
                          alpha=min_limit,
                          shift=0,
                          scale=1.0)

        builder.add_unary(name=node.name + '_min_minus_x_minus_b',
                          input_name=node.inputs[0] + '_min_x_a',
                          output_name=node.inputs[0] + '_min_minus_x_minus_b',
                          mode='threshold',
                          alpha=-max_limit,
                          shift=0,
                          scale=-1.0)

        builder.add_activation(name=node.name,
                               non_linearity='LINEAR',
                               input_name=node.inputs[0] + '_min_minus_x_minus_b',
                               output_name=node.outputs[0],
                               params=[-1.0, 0])

    _update_shape_mapping_unchanged(node, graph, err)

def _convert_mvn(builder, node, graph, err): # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_mvn(name=node.name,
                    input_name=node.inputs[0],
                    output_name=node.outputs[0],
                    across_channels = node.attrs.get('across_channels', 0),
                    normalize_variance = node.attrs.get('normalize_variance', 1),
                    epsilon = 1e-5)
    _update_shape_mapping_unchanged(node, graph, err)

def _convert_lstm(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    W_name = node.inputs[1]
    R_name = node.inputs[2]
    B = None
    if len(node.inputs) > 3:
        B_name = node.inputs[3]
        B = node.input_tensors.get(B_name, None)
    W = node.input_tensors.get(W_name, None)
    R = node.input_tensors.get(R_name, None)
    if W is None:
        err.missing_initializer(node,
                                "Weight tensor: {} not found in the graph initializer".format(W_name, ))
    if R is None:
        err.missing_initializer(node,
                                "Weight tensor: {} not found in the graph initializer".format(R_name, ))

    h = node.attrs["hidden_size"]
    W_i, W_o, W_f, W_c = np.split(np.squeeze(W), 4)  #type: ignore
    R_i, R_o, R_f, R_c = np.split(np.squeeze(R), 4)  #type: ignore
    x = W_i.shape[1]
    h = W_i.shape[0]
    W_x = [W_i, W_f, W_o, W_c]
    W_h = [R_i, R_f, R_o, R_c]
    b = None
    if B is not None:
        b_Wi, b_Wo, b_Wf, b_Wc, b_Ri, b_Ro, b_Rf, b_Rc = np.split(np.squeeze(B), 8)  #type: ignore
        b = [b_Wi + b_Ri, b_Wf + b_Rf, b_Wo + b_Ro, b_Wc + b_Rc]

    input_h = node.inputs[5] if len(node.inputs) > 5 else node.inputs[0] + '_h_input'
    input_c = node.inputs[6] if len(node.inputs) > 6 else node.inputs[0] + '_c_input'
    output_h = node.outputs[1] if len(node.outputs) > 1 else node.outputs[0] + '_h_output'
    output_c = node.outputs[2] if len(node.outputs) > 2 else node.outputs[0] + '_c_output'

    graph.optional_inputs.append((input_h, (h)))
    graph.optional_inputs.append((input_c, (h)))
    graph.optional_outputs.append((output_h, (h)))
    graph.optional_outputs.append((output_c, (h)))

    builder.add_unilstm(name = node.name,
                    W_h = W_h,
                    W_x = W_x,
                    b = b,
                    hidden_size = h,
                    input_size = x,
                    input_names= [node.inputs[0], input_h, input_c],
                    output_names= [node.outputs[0], output_h, output_c],
                    inner_activation='SIGMOID',
                    cell_state_update_activation='TANH',
                    output_activation='TANH',
                    peep=None,
                    output_all=True,
                    forget_bias=False, coupled_input_forget_gate=False,
                    cell_clip_threshold=50000.0, reverse_input=False)

    if _is_input_shape_mapping_defined(node, graph):
        graph.onnx_coreml_shape_mapping[node.outputs[0]] = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        graph.onnx_coreml_shape_mapping[node.outputs[1]] = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        graph.onnx_coreml_shape_mapping[node.outputs[2]] = graph.onnx_coreml_shape_mapping[node.inputs[0]]



def _convert_custom(builder, node, graph, err): # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    params = NeuralNetwork_pb2.CustomLayerParams()
    params.className = node.op_type
    params.description = "Custom layer that corresponds to the ONNX op {}".format(node.op_type,)

    inputs_ = []
    # skip the inputs that are initializers
    for inp in node.inputs:
        if inp not in node.input_tensors:
            inputs_.append(inp)

    builder.add_custom(
        name=node.name,
        input_names=inputs_,
        output_names=node.outputs,
        custom_proto_spec=params
    )
    err.custom_layer_nodes.append(node)

def _convert_identity(builder, node, graph, err): # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    builder.add_activation(
        name=node.name,
        non_linearity = 'LINEAR',
        input_name=node.inputs[0],
        output_name=node.outputs[0],
        params=[1.0, 0.0]
    )
    if _is_input_shape_mapping_defined(node, graph):
        mapp = graph.onnx_coreml_shape_mapping[node.inputs[0]]
        mapp_out = []
        if node.op_type == 'Squeeze':
            axes = node.attrs.get('axes', None)
            if axes is None:
                if node.inputs[0] not in graph.shape_dict:
                    return err.unsupported_op_configuration(builder, node, graph, "shape not known")
                else:
                    ishape = graph.shape_dict[node.inputs[0]]
                    if ishape.count(1) == len(ishape):
                        mapp_out = [2]
                    else:
                        for i, d in enumerate(ishape):
                            if d != 1:
                                mapp_out.append(mapp[i])
            else:
                for i, a in enumerate(mapp):
                    if i in axes:
                        continue
                    else:
                        mapp_out.append(a)
                if len(mapp_out) == 0:
                    mapp_out = [2]
        elif node.op_type == 'Unsqueeze':
            axes = node.attrs['axes']
            available_set = [0,1,2,3,4]
            for d in mapp:
                if d in available_set:
                    available_set.remove(d)
            if len(axes) > len(available_set):
                return err.unsupported_op_configuration(builder, node, graph, "cannot unsqueeze to a dimension greater than 5")
            mapp_out = [1] * (len(axes) + len(mapp))
            mapp_ptr = 0
            available_set_ptr = 0
            for i in range(len(mapp_out)):
                if i in axes:
                    mapp_out[i] = available_set[available_set_ptr]
                    available_set_ptr += 1
                else:
                    mapp_out[i] = mapp[mapp_ptr]
                    mapp_ptr += 1
        else:
            raise ValueError('convert_identity incorrectly called')
        graph.onnx_coreml_shape_mapping[node.outputs[0]] = mapp_out


def _convert_const(builder, node, graph, err): # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None

    mapp = None
    for input_ in node.inputs:
        if input_ in graph.onnx_coreml_shape_mapping:
            mapp = graph.onnx_coreml_shape_mapping[input_]

    for name, value in node.input_tensors.items():
        output_name = name
        if name not in graph.constant_layers_added:
            add_transpose_later = False
            shape = value.shape
            coreml_shape = [1,1,1]
            if len(shape) == 0:
                graph.onnx_coreml_shape_mapping[name] = [2]  # [C]
            elif len(shape) == 3:
                coreml_shape = list(shape)
                graph.onnx_coreml_shape_mapping[name] = [2, 3, 4] # [C,H,W]
            elif len(shape) == 1:
                coreml_shape = [shape[0],1,1]
                graph.onnx_coreml_shape_mapping[name] = [2]  # [C]
            elif len(shape) == 2:
                coreml_shape = [1, shape[0], shape[1]]
                if mapp is not None and (mapp == [1,2] or mapp == [0,2]):
                    add_transpose_later = True
                    transpose_dims = [2,3,0,1]
                    graph.onnx_coreml_shape_mapping[name] = [0, 2]  # [S,C]
                else:
                    graph.onnx_coreml_shape_mapping[name] = [3, 4]  # [H,W]
            else:
                return err.unsupported_op_configuration(builder, node, graph, "unable to translate constant array shape to CoreML shape")

            if add_transpose_later:
                output_name += '_pre_transpose'
            builder.add_load_constant(name=output_name,
                                  output_name=output_name,
                                  constant_value=value.flatten(),
                                  shape=coreml_shape)
            if add_transpose_later:
                builder.add_permute(
                    name=name,
                    dim=transpose_dims,
                    input_name=output_name,
                    output_name=name
                )

            graph.constant_layers_added[output_name] = True


_ONNX_NODE_REGISTRY = {
    "Abs": _convert_abs,
    "Add": _convert_add,
    "ArgMax": _convert_argmax,
    "ArgMin": _convert_argmax,
    "AveragePool": _convert_pool,
    "BatchNormalization": _convert_bn,
    "Clip": _convert_clip,
    "Concat": _convert_concat,
    "Conv": _convert_conv,
    "ConvTranspose": _convert_conv,
    "DepthToSpace": _convert_reorganize_data,
    "Div": _convert_div,
    "Elu": _convert_elu,
    "Exp": _convert_exp,
    "Flatten": _convert_flatten, #Todo: handle more cases
    "Gemm": _convert_gemm,
    "GlobalAveragePool": _convert_pool,
    "GlobalMaxPool": _convert_pool,
    "HardSigmoid": _convert_hardsigmoid,
    "InstanceNormalization": _convert_instancenorm,
    "LeakyRelu": _convert_leaky_relu,
    "Log": _convert_log,
    "LogSoftmax": _convert_softmax,
    "LRN": _convert_lrn,
    "LSTM": _convert_lstm,
    "MatMul": _convert_matmul,
    "Max": _convert_max,
    "MaxPool": _convert_pool,
    "Mean": _convert_mean,
    "MeanVarianceNormalization": _convert_mvn,
    "Min": _convert_min,
    "Mul": _convert_mul,
    "Neg": _convert_neg,
    "Pad": _convert_pad,
    "Pow": _convert_pow,
    "PRelu": _convert_prelu,
    "Reciprocal": _convert_reciprocal,
    "ReduceL1": _convert_reduce,
    "ReduceL2": _convert_reduce,
    "ReduceLogSum": _convert_reduce,
    "ReduceMax": _convert_reduce,
    "ReduceMean": _convert_reduce,
    "ReduceMin": _convert_reduce,
    "ReduceProd": _convert_reduce,
    "ReduceSum": _convert_reduce,
    "ReduceSumSquare": _convert_reduce,
    "Relu": _convert_relu,
    "Reshape": _convert_reshape,
    "Selu": _convert_selu,
    "Sigmoid": _convert_sigmoid,
    "Sign": _convert_sign,
    "Slice": _convert_slice,
    "Softmax": _convert_softmax, #Todo: handle more cases
    "Softplus": _convert_softplus,
    "Softsign": _convert_softsign,
    "SpaceToDepth": _convert_reorganize_data,
    "SpatialBN": _convert_bn,
    "Split": _convert_split,
    "Sqrt": _convert_sqrt,
    "Squeeze": _convert_identity,
    "Sub": _convert_sub,
    "Sum": _convert_add,
    "Tanh": _convert_tanh,
    "ThresholdedRelu": _convert_thresholdedrelu,
    "Transpose": _convert_transpose,
    "Unsqueeze": _convert_identity,
    "Upsample": _convert_upsample,
}

_SEQUENCE_LAYERS_REGISTRY = set(["LSTM"])

_CONST_INPUT_ALLOWED_LAYERS = set([ "Add", "Sum", "Mul", "Concat", "Max", "Min", "Div", "Reciprocal"])

def _get_node_converter_fn(builder, node, err):  # type: (NeuralNetworkBuilder, Node, ErrorHandling) -> Callable[[NeuralNetworkBuilder, Node, Graph, ErrorHandling], None]
    """
    Get the right converter function for ONNX node op_type
    """
    op_type = node.op_type
    # Return custom conversion function if provided
    # If both node type and node name custom function
    # is provided, then use node name specific custom function, as
    # type specific custom function is more generic than name specific
    if node.name in err.custom_conversion_functions:
        return err.custom_conversion_functions[node.name]
    elif op_type in err.custom_conversion_functions:
        return err.custom_conversion_functions[op_type]
    elif op_type in _ONNX_NODE_REGISTRY:
        return _ONNX_NODE_REGISTRY[op_type]
    else:
        return err.unsupported_op(node)

def _add_const_inputs_if_required(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    if node.op_type in _CONST_INPUT_ALLOWED_LAYERS:
        if len(node.input_tensors) > 0:
            _convert_const(builder, node, graph, err)


def _convert_node(builder, node, graph, err):  # type: (NeuralNetworkBuilder, Node, Graph, ErrorHandling) -> None
    converter_fn = _get_node_converter_fn(builder, node, err)
    return converter_fn(builder, node, graph, err)