RuntimeError: Exporting the operator var to ONNX opset version 11 is not supported. Please open a bu

jetson tx2 部署.pth文件–onnx时报错：
RuntimeError: Exporting the operator var to ONNX opset version 11 is not supported. Please open a bug to request ONNX export support for the missing operator.
是因为torch.var不支持。
解决办法：
在/anaconda3/envs/torch1.7/lib/python3.7/site-packages/torch/onnx/symbolic_opset11.py中加入以下代码段，即可顺利解决。

# This file exports ONNX ops for opset 11
import functools
import math
import sys
import warnings
from typing import List, Optional, Tuple, Union

import torch
import torch._C._onnx as _C_onnx
import torch.nn.modules.utils
import torch.onnx
from torch import _C

# Monkey-patch graph manipulation methods on Graph, used for the ONNX symbolics
from torch.onnx import symbolic_helper


@symbolic_helper.parse_args("v", "is", "i", "i")
def _var_mean(g, input, dim, correction, keepdim):
    if dim is None:
        mean = g.op("ReduceMean", input, keepdims_i=0)
        t_mean = mean
        num_elements = numel(g, input)
    else:
        mean = g.op("ReduceMean", input, axes_i=dim, keepdims_i=keepdim)
        t_mean = g.op("ReduceMean", input, axes_i=dim, keepdims_i=1)
        redudced_dims = g.op("Shape", input)
        # dim could contain one or multiple dimensions
        redudced_dims = g.op(
            "Gather",
            redudced_dims,
            g.op("Constant", value_t=torch.tensor(dim)),
            axis_i=0,
        )
        num_elements = g.op("ReduceProd", redudced_dims, keepdims_i=0)
    sub_v = g.op("Sub", input, t_mean)
    sqr_sub = g.op("Mul", sub_v, sub_v)
    keepdim_mean = 0 if dim is None else keepdim
    var = g.op("ReduceMean", sqr_sub, axes_i=dim, keepdims_i=keepdim_mean)
    # Correct bias in calculating variance, by dividing it over (N - correction) instead on N
    if correction is None:
        correction = 1
    if correction != 0:
        num_elements = g.op(
            "Cast", num_elements, to_i=symbolic_helper.cast_pytorch_to_onnx["Float"]
        )
        one = g.op("Constant", value_t=torch.tensor(correction, dtype=torch.float))
        mul = g.op("Mul", var, num_elements)
        var = g.op("Div", mul, g.op("Sub", num_elements, one))
    return var, mean


def std(g, input, *args):
    var, _ = var_mean(g, input, *args)
    return g.op("Sqrt", var)


def var(g, input, *args):
    var, _ = var_mean(g, input, *args)
    return var

# var_mean (and all variance-related functions) has multiple signatures, so need to manually figure
# out the correct arguments:
# aten::var_mean(Tensor self, bool unbiased)
# aten::var_mean(Tensor self, int[1] dim, bool unbiased, bool keepdim=False)
# aten::var_mean(Tensor self, int[1]? dim=None, *, int? correction=None, bool keepdim=False)
def var_mean(g, input, *args):
    if len(args) == 1:
        return _var_mean(g, input, None, args[0], None)
    else:
        return _var_mean(g, input, *args)


def std_mean(g, input, *args):
    var, mean = var_mean(g, input, *args)
    return g.op("Sqrt", var), mean