How to use the mxnet.symbol.BatchNorm function in mxnet

data_tmp = np.random.normal(-0.1, 0.1, size=shape)
            s = shape[1],
            gamma = np.ones(s)
            beta = np.ones(s)
            gamma[1] = 3
            beta[0] = 3

            rolling_mean = np.random.uniform(size=s)
            rolling_std = np.random.uniform(size=s)

            data = mx.symbol.Variable('data', stype=stype)
            in_location = [mx.nd.array(data_tmp).tostype(stype), mx.nd.array(gamma).tostype(stype),
                           mx.nd.array(beta).tostype(stype)]
            mean_std = [mx.nd.array(rolling_mean).tostype(stype), mx.nd.array(rolling_std).tostype(stype)]

            test = mx.symbol.BatchNorm(data, fix_gamma=False)
            check_numeric_gradient(test, in_location, mean_std, numeric_eps=1e-2, rtol=0.16, atol=1e-2)

pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b4_branch2c = mx.symbol.BatchNorm(name='bn4b4_branch2c', data=res4b4_branch2c,
                                             use_global_stats=self.use_global_stats, eps=self.eps, fix_gamma=False)
        scale4b4_branch2c = bn4b4_branch2c
        res4b4 = mx.symbol.broadcast_add(name='res4b4', *[res4b3_relu, scale4b4_branch2c])
        res4b4_relu = mx.symbol.Activation(name='res4b4_relu', data=res4b4, act_type='relu')
        res4b5_branch2a = mx.symbol.Convolution(name='res4b5_branch2a', data=res4b4_relu, num_filter=256, pad=(0, 0),
                                                kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b5_branch2a = mx.symbol.BatchNorm(name='bn4b5_branch2a', data=res4b5_branch2a,
                                             use_global_stats=self.use_global_stats, eps=self.eps, fix_gamma=False)
        scale4b5_branch2a = bn4b5_branch2a
        res4b5_branch2a_relu = mx.symbol.Activation(name='res4b5_branch2a_relu', data=scale4b5_branch2a,
                                                    act_type='relu')
        res4b5_branch2b = mx.symbol.Convolution(name='res4b5_branch2b', data=res4b5_branch2a_relu, num_filter=256,
                                                pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn4b5_branch2b = mx.symbol.BatchNorm(name='bn4b5_branch2b', data=res4b5_branch2b,
                                             use_global_stats=self.use_global_stats, eps=self.eps, fix_gamma=False)
        scale4b5_branch2b = bn4b5_branch2b
        res4b5_branch2b_relu = mx.symbol.Activation(name='res4b5_branch2b_relu', data=scale4b5_branch2b,
                                                    act_type='relu')
        res4b5_branch2c = mx.symbol.Convolution(name='res4b5_branch2c', data=res4b5_branch2b_relu, num_filter=1024,
                                                pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b5_branch2c = mx.symbol.BatchNorm(name='bn4b5_branch2c', data=res4b5_branch2c,
                                             use_global_stats=self.use_global_stats, eps=self.eps, fix_gamma=False)
        scale4b5_branch2c = bn4b5_branch2c
        res4b5 = mx.symbol.broadcast_add(name='res4b5', *[res4b4_relu, scale4b5_branch2c])
        res4b5_relu = mx.symbol.Activation(name='res4b5_relu', data=res4b5, act_type='relu')
        res4b6_branch2a = mx.symbol.Convolution(name='res4b6_branch2a', data=res4b5_relu, num_filter=256, pad=(0, 0),
                                                kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b6_branch2a = mx.symbol.BatchNorm(name='bn4b6_branch2a', data=res4b6_branch2a,
                                             use_global_stats=self.use_global_stats, eps=self.eps, fix_gamma=False)
        scale4b6_branch2a = bn4b6_branch2a

res3a_branch2b = mx.symbol.Convolution(name='res3a_branch2b', data=res3a_branch2a_relu, num_filter=128, pad=(1, 1),
                                               kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn3a_branch2b = mx.symbol.BatchNorm(name='bn3a_branch2b', data=res3a_branch2b, use_global_stats=True,
                                            fix_gamma=False, eps=self.eps)
        scale3a_branch2b = bn3a_branch2b
        res3a_branch2b_relu = mx.symbol.Activation(name='res3a_branch2b_relu', data=scale3a_branch2b, act_type='relu')
        res3a_branch2c = mx.symbol.Convolution(name='res3a_branch2c', data=res3a_branch2b_relu, num_filter=512, pad=(0, 0),
                                               kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn3a_branch2c = mx.symbol.BatchNorm(name='bn3a_branch2c', data=res3a_branch2c, use_global_stats=True,
                                            fix_gamma=False, eps=self.eps)
        scale3a_branch2c = bn3a_branch2c
        res3a = mx.symbol.broadcast_add(name='res3a', *[scale3a_branch1, scale3a_branch2c])
        res3a_relu = mx.symbol.Activation(name='res3a_relu', data=res3a, act_type='relu')
        res3b1_branch2a = mx.symbol.Convolution(name='res3b1_branch2a', data=res3a_relu, num_filter=128, pad=(0, 0),
                                                kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn3b1_branch2a = mx.symbol.BatchNorm(name='bn3b1_branch2a', data=res3b1_branch2a, use_global_stats=True,
                                             fix_gamma=False, eps=self.eps)
        scale3b1_branch2a = bn3b1_branch2a
        res3b1_branch2a_relu = mx.symbol.Activation(name='res3b1_branch2a_relu', data=scale3b1_branch2a, act_type='relu')
        res3b1_branch2b = mx.symbol.Convolution(name='res3b1_branch2b', data=res3b1_branch2a_relu, num_filter=128,
                                                pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn3b1_branch2b = mx.symbol.BatchNorm(name='bn3b1_branch2b', data=res3b1_branch2b, use_global_stats=True,
                                             fix_gamma=False, eps=self.eps)
        scale3b1_branch2b = bn3b1_branch2b
        res3b1_branch2b_relu = mx.symbol.Activation(name='res3b1_branch2b_relu', data=scale3b1_branch2b, act_type='relu')
        res3b1_branch2c = mx.symbol.Convolution(name='res3b1_branch2c', data=res3b1_branch2b_relu, num_filter=512,
                                                pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn3b1_branch2c = mx.symbol.BatchNorm(name='bn3b1_branch2c', data=res3b1_branch2c, use_global_stats=True,
                                             fix_gamma=False, eps=self.eps)
        scale3b1_branch2c = bn3b1_branch2c
        res3b1 = mx.symbol.broadcast_add(name='res3b1', *[res3a_relu, scale3b1_branch2c])
        res3b1_relu = mx.symbol.Activation(name='res3b1_relu', data=res3b1, act_type='relu')

pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b9_branch2c = mx.symbol.BatchNorm(name='bn4b9_branch2c', data=res4b9_branch2c, use_global_stats=True,
                                             fix_gamma=False, eps=eps)
        scale4b9_branch2c = bn4b9_branch2c
        res4b9 = mx.symbol.broadcast_add(name='res4b9', *[res4b8_relu, scale4b9_branch2c])
        res4b9_relu = mx.symbol.Activation(name='res4b9_relu', data=res4b9, act_type='relu')
        res4b10_branch2a = mx.symbol.Convolution(name='res4b10_branch2a', data=res4b9_relu, num_filter=256, pad=(0, 0),
                                                 kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b10_branch2a = mx.symbol.BatchNorm(name='bn4b10_branch2a', data=res4b10_branch2a, use_global_stats=True,
                                              fix_gamma=False, eps=eps)
        scale4b10_branch2a = bn4b10_branch2a
        res4b10_branch2a_relu = mx.symbol.Activation(name='res4b10_branch2a_relu', data=scale4b10_branch2a,
                                                     act_type='relu')
        res4b10_branch2b = mx.symbol.Convolution(name='res4b10_branch2b', data=res4b10_branch2a_relu, num_filter=256,
                                                 pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn4b10_branch2b = mx.symbol.BatchNorm(name='bn4b10_branch2b', data=res4b10_branch2b, use_global_stats=True,
                                              fix_gamma=False, eps=eps)
        scale4b10_branch2b = bn4b10_branch2b
        res4b10_branch2b_relu = mx.symbol.Activation(name='res4b10_branch2b_relu', data=scale4b10_branch2b,
                                                     act_type='relu')
        res4b10_branch2c = mx.symbol.Convolution(name='res4b10_branch2c', data=res4b10_branch2b_relu, num_filter=1024,
                                                 pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b10_branch2c = mx.symbol.BatchNorm(name='bn4b10_branch2c', data=res4b10_branch2c, use_global_stats=True,
                                              fix_gamma=False, eps=eps)
        scale4b10_branch2c = bn4b10_branch2c
        res4b10 = mx.symbol.broadcast_add(name='res4b10', *[res4b9_relu, scale4b10_branch2c])
        res4b10_relu = mx.symbol.Activation(name='res4b10_relu', data=res4b10, act_type='relu')
        res4b11_branch2a = mx.symbol.Convolution(name='res4b11_branch2a', data=res4b10_relu, num_filter=256, pad=(0, 0),
                                                 kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b11_branch2a = mx.symbol.BatchNorm(name='bn4b11_branch2a', data=res4b11_branch2a, use_global_stats=True,
                                              fix_gamma=False, eps=eps)
        scale4b11_branch2a = bn4b11_branch2a

res3b2_relu = mx.symbol.Activation(name='res3b2_relu', data=res3b2, act_type='relu')
        res3b3_branch2a = mx.symbol.Convolution(name='res3b3_branch2a', data=res3b2_relu, num_filter=128, pad=(0, 0),
                                                kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn3b3_branch2a = mx.symbol.BatchNorm(name='bn3b3_branch2a', data=res3b3_branch2a, use_global_stats=True,
                                             fix_gamma=False, eps=self.eps)
        scale3b3_branch2a = bn3b3_branch2a
        res3b3_branch2a_relu = mx.symbol.Activation(name='res3b3_branch2a_relu', data=scale3b3_branch2a, act_type='relu')
        res3b3_branch2b = mx.symbol.Convolution(name='res3b3_branch2b', data=res3b3_branch2a_relu, num_filter=128,
                                                pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn3b3_branch2b = mx.symbol.BatchNorm(name='bn3b3_branch2b', data=res3b3_branch2b, use_global_stats=True,
                                             fix_gamma=False, eps=self.eps)
        scale3b3_branch2b = bn3b3_branch2b
        res3b3_branch2b_relu = mx.symbol.Activation(name='res3b3_branch2b_relu', data=scale3b3_branch2b, act_type='relu')
        res3b3_branch2c = mx.symbol.Convolution(name='res3b3_branch2c', data=res3b3_branch2b_relu, num_filter=512,
                                                pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn3b3_branch2c = mx.symbol.BatchNorm(name='bn3b3_branch2c', data=res3b3_branch2c, use_global_stats=True,
                                             fix_gamma=False, eps=self.eps)
        scale3b3_branch2c = bn3b3_branch2c
        res3b3 = mx.symbol.broadcast_add(name='res3b3', *[res3b2_relu, scale3b3_branch2c])
        res3b3_relu = mx.symbol.Activation(name='res3b3_relu', data=res3b3, act_type='relu')
        res4a_branch1 = mx.symbol.Convolution(name='res4a_branch1', data=res3b3_relu, num_filter=1024, pad=(0, 0),
                                              kernel=(1, 1), stride=(2, 2), no_bias=True)
        bn4a_branch1 = mx.symbol.BatchNorm(name='bn4a_branch1', data=res4a_branch1, use_global_stats=True, fix_gamma=False, eps=self.eps)
        scale4a_branch1 = bn4a_branch1
        res4a_branch2a = mx.symbol.Convolution(name='res4a_branch2a', data=res3b3_relu, num_filter=256, pad=(0, 0),
                                               kernel=(1, 1), stride=(2, 2), no_bias=True)
        bn4a_branch2a = mx.symbol.BatchNorm(name='bn4a_branch2a', data=res4a_branch2a, use_global_stats=True,
                                            fix_gamma=False, eps=self.eps)
        scale4a_branch2a = bn4a_branch2a
        res4a_branch2a_relu = mx.symbol.Activation(name='res4a_branch2a_relu', data=scale4a_branch2a, act_type='relu')
        res4a_branch2b = mx.symbol.Convolution(name='res4a_branch2b', data=res4a_branch2a_relu, num_filter=256, pad=(1, 1),
                                               kernel=(3, 3), stride=(1, 1), no_bias=True)

res3b3_branch2b = mx.contrib.symbol.DeformableConvolution(name='res3b3_branch2b', data=res3b3_branch2a_relu,
                                                                      offset=res3b3_branch2b_offset,
                                                                      num_filter=128, pad=(1, 1), kernel=(3, 3),
                                                                      num_deformable_group=4,
                                                                      stride=(1, 1), no_bias=True)
        else:
            res3b3_branch2b = mx.symbol.Convolution(name='res3b3_branch2b', data=res3b3_branch2a_relu, num_filter=128,
                                                    pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn3b3_branch2b = mx.symbol.BatchNorm(name='bn3b3_branch2b', data=res3b3_branch2b, use_global_stats=True,
                                             fix_gamma=False, eps=eps)
        scale3b3_branch2b = bn3b3_branch2b
        res3b3_branch2b_relu = mx.symbol.Activation(name='res3b3_branch2b_relu', data=scale3b3_branch2b,
                                                    act_type='relu')
        res3b3_branch2c = mx.symbol.Convolution(name='res3b3_branch2c', data=res3b3_branch2b_relu, num_filter=512,
                                                pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn3b3_branch2c = mx.symbol.BatchNorm(name='bn3b3_branch2c', data=res3b3_branch2c, use_global_stats=True,
                                             fix_gamma=False, eps=eps)
        scale3b3_branch2c = bn3b3_branch2c
        res3b3 = mx.symbol.broadcast_add(name='res3b3', *[res3b2_relu, scale3b3_branch2c])
        res3b3_relu = mx.symbol.Activation(name='res3b3_relu', data=res3b3, act_type='relu')
        res4a_branch1 = mx.symbol.Convolution(name='res4a_branch1', data=res3b3_relu, num_filter=1024, pad=(0, 0),
                                              kernel=(1, 1), stride=(2, 2), no_bias=True)
        bn4a_branch1 = mx.symbol.BatchNorm(name='bn4a_branch1', data=res4a_branch1, use_global_stats=True,
                                           fix_gamma=False, eps=eps)
        scale4a_branch1 = bn4a_branch1
        res4a_branch2a = mx.symbol.Convolution(name='res4a_branch2a', data=res3b3_relu, num_filter=256, pad=(0, 0),
                                               kernel=(1, 1), stride=(2, 2), no_bias=True)
        bn4a_branch2a = mx.symbol.BatchNorm(name='bn4a_branch2a', data=res4a_branch2a, use_global_stats=True,
                                            fix_gamma=False, eps=eps)
        scale4a_branch2a = bn4a_branch2a
        res4a_branch2a_relu = mx.symbol.Activation(name='res4a_branch2a_relu', data=scale4a_branch2a, act_type='relu')
        res4a_branch2b = mx.symbol.Convolution(name='res4a_branch2b', data=res4a_branch2a_relu, num_filter=256,

res4b10_branch2b = mx.symbol.Convolution(name='res4b10_branch2b', data=res4b10_branch2a_relu, num_filter=256,
                                                 pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn4b10_branch2b = mx.symbol.BatchNorm(name='bn4b10_branch2b', data=res4b10_branch2b, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b10_branch2b = bn4b10_branch2b
        res4b10_branch2b_relu = mx.symbol.Activation(name='res4b10_branch2b_relu', data=scale4b10_branch2b, act_type='relu')
        res4b10_branch2c = mx.symbol.Convolution(name='res4b10_branch2c', data=res4b10_branch2b_relu, num_filter=1024,
                                                 pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b10_branch2c = mx.symbol.BatchNorm(name='bn4b10_branch2c', data=res4b10_branch2c, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b10_branch2c = bn4b10_branch2c
        res4b10 = mx.symbol.broadcast_add(name='res4b10', *[res4b9_relu, scale4b10_branch2c])
        res4b10_relu = mx.symbol.Activation(name='res4b10_relu', data=res4b10, act_type='relu')
        res4b11_branch2a = mx.symbol.Convolution(name='res4b11_branch2a', data=res4b10_relu, num_filter=256, pad=(0, 0),
                                                 kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b11_branch2a = mx.symbol.BatchNorm(name='bn4b11_branch2a', data=res4b11_branch2a, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b11_branch2a = bn4b11_branch2a
        res4b11_branch2a_relu = mx.symbol.Activation(name='res4b11_branch2a_relu', data=scale4b11_branch2a, act_type='relu')
        res4b11_branch2b = mx.symbol.Convolution(name='res4b11_branch2b', data=res4b11_branch2a_relu, num_filter=256,
                                                 pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn4b11_branch2b = mx.symbol.BatchNorm(name='bn4b11_branch2b', data=res4b11_branch2b, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b11_branch2b = bn4b11_branch2b
        res4b11_branch2b_relu = mx.symbol.Activation(name='res4b11_branch2b_relu', data=scale4b11_branch2b, act_type='relu')
        res4b11_branch2c = mx.symbol.Convolution(name='res4b11_branch2c', data=res4b11_branch2b_relu, num_filter=1024,
                                                 pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b11_branch2c = mx.symbol.BatchNorm(name='bn4b11_branch2c', data=res4b11_branch2c, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b11_branch2c = bn4b11_branch2c
        res4b11 = mx.symbol.broadcast_add(name='res4b11', *[res4b10_relu, scale4b11_branch2c])
        res4b11_relu = mx.symbol.Activation(name='res4b11_relu', data=res4b11, act_type='relu')

res4b11_branch2b = mx.symbol.Convolution(name='res4b11_branch2b', data=res4b11_branch2a_relu, num_filter=256,
                                                 pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn4b11_branch2b = mx.symbol.BatchNorm(name='bn4b11_branch2b', data=res4b11_branch2b, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b11_branch2b = bn4b11_branch2b
        res4b11_branch2b_relu = mx.symbol.Activation(name='res4b11_branch2b_relu', data=scale4b11_branch2b, act_type='relu')
        res4b11_branch2c = mx.symbol.Convolution(name='res4b11_branch2c', data=res4b11_branch2b_relu, num_filter=1024,
                                                 pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b11_branch2c = mx.symbol.BatchNorm(name='bn4b11_branch2c', data=res4b11_branch2c, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b11_branch2c = bn4b11_branch2c
        res4b11 = mx.symbol.broadcast_add(name='res4b11', *[res4b10_relu, scale4b11_branch2c])
        res4b11_relu = mx.symbol.Activation(name='res4b11_relu', data=res4b11, act_type='relu')
        res4b12_branch2a = mx.symbol.Convolution(name='res4b12_branch2a', data=res4b11_relu, num_filter=256, pad=(0, 0),
                                                 kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b12_branch2a = mx.symbol.BatchNorm(name='bn4b12_branch2a', data=res4b12_branch2a, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b12_branch2a = bn4b12_branch2a
        res4b12_branch2a_relu = mx.symbol.Activation(name='res4b12_branch2a_relu', data=scale4b12_branch2a, act_type='relu')
        res4b12_branch2b = mx.symbol.Convolution(name='res4b12_branch2b', data=res4b12_branch2a_relu, num_filter=256,
                                                 pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn4b12_branch2b = mx.symbol.BatchNorm(name='bn4b12_branch2b', data=res4b12_branch2b, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b12_branch2b = bn4b12_branch2b
        res4b12_branch2b_relu = mx.symbol.Activation(name='res4b12_branch2b_relu', data=scale4b12_branch2b, act_type='relu')
        res4b12_branch2c = mx.symbol.Convolution(name='res4b12_branch2c', data=res4b12_branch2b_relu, num_filter=1024,
                                                 pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b12_branch2c = mx.symbol.BatchNorm(name='bn4b12_branch2c', data=res4b12_branch2c, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b12_branch2c = bn4b12_branch2c
        res4b12 = mx.symbol.broadcast_add(name='res4b12', *[res4b11_relu, scale4b12_branch2c])
        res4b12_relu = mx.symbol.Activation(name='res4b12_relu', data=res4b12, act_type='relu')

res4b19_relu = mx.symbol.Activation(name='res4b19_relu', data=res4b19, act_type='relu')
        res4b20_branch2a = mx.symbol.Convolution(name='res4b20_branch2a', data=res4b19_relu, num_filter=256, pad=(0, 0),
                                                 kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b20_branch2a = mx.symbol.BatchNorm(name='bn4b20_branch2a', data=res4b20_branch2a, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b20_branch2a = bn4b20_branch2a
        res4b20_branch2a_relu = mx.symbol.Activation(name='res4b20_branch2a_relu', data=scale4b20_branch2a, act_type='relu')
        res4b20_branch2b = mx.symbol.Convolution(name='res4b20_branch2b', data=res4b20_branch2a_relu, num_filter=256,
                                                 pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn4b20_branch2b = mx.symbol.BatchNorm(name='bn4b20_branch2b', data=res4b20_branch2b, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b20_branch2b = bn4b20_branch2b
        res4b20_branch2b_relu = mx.symbol.Activation(name='res4b20_branch2b_relu', data=scale4b20_branch2b, act_type='relu')
        res4b20_branch2c = mx.symbol.Convolution(name='res4b20_branch2c', data=res4b20_branch2b_relu, num_filter=1024,
                                                 pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b20_branch2c = mx.symbol.BatchNorm(name='bn4b20_branch2c', data=res4b20_branch2c, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b20_branch2c = bn4b20_branch2c
        res4b20 = mx.symbol.broadcast_add(name='res4b20', *[res4b19_relu, scale4b20_branch2c])
        res4b20_relu = mx.symbol.Activation(name='res4b20_relu', data=res4b20, act_type='relu')
        res4b21_branch2a = mx.symbol.Convolution(name='res4b21_branch2a', data=res4b20_relu, num_filter=256, pad=(0, 0),
                                                 kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b21_branch2a = mx.symbol.BatchNorm(name='bn4b21_branch2a', data=res4b21_branch2a, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b21_branch2a = bn4b21_branch2a
        res4b21_branch2a_relu = mx.symbol.Activation(name='res4b21_branch2a_relu', data=scale4b21_branch2a, act_type='relu')
        res4b21_branch2b = mx.symbol.Convolution(name='res4b21_branch2b', data=res4b21_branch2a_relu, num_filter=256,
                                                 pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn4b21_branch2b = mx.symbol.BatchNorm(name='bn4b21_branch2b', data=res4b21_branch2b, use_global_stats=True,
                                              fix_gamma=False, eps=self.eps)
        scale4b21_branch2b = bn4b21_branch2b
        res4b21_branch2b_relu = mx.symbol.Activation(name='res4b21_branch2b_relu', data=scale4b21_branch2b, act_type='relu')

pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b12_branch2c = mx.symbol.BatchNorm(name='bn4b12_branch2c', data=res4b12_branch2c, use_global_stats=True,
                                              fix_gamma=False, eps = self.eps)
        scale4b12_branch2c = bn4b12_branch2c
        res4b12 = mx.symbol.broadcast_add(name='res4b12', *[res4b11_relu, scale4b12_branch2c])
        res4b12_relu = mx.symbol.Activation(name='res4b12_relu', data=res4b12, act_type='relu')
        res4b13_branch2a = mx.symbol.Convolution(name='res4b13_branch2a', data=res4b12_relu, num_filter=256, pad=(0, 0),
                                                 kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b13_branch2a = mx.symbol.BatchNorm(name='bn4b13_branch2a', data=res4b13_branch2a, use_global_stats=True,
                                              fix_gamma=False, eps = self.eps)
        scale4b13_branch2a = bn4b13_branch2a
        res4b13_branch2a_relu = mx.symbol.Activation(name='res4b13_branch2a_relu', data=scale4b13_branch2a,
                                                     act_type='relu')
        res4b13_branch2b = mx.symbol.Convolution(name='res4b13_branch2b', data=res4b13_branch2a_relu, num_filter=256,
                                                 pad=(1, 1), kernel=(3, 3), stride=(1, 1), no_bias=True)
        bn4b13_branch2b = mx.symbol.BatchNorm(name='bn4b13_branch2b', data=res4b13_branch2b, use_global_stats=True,
                                              fix_gamma=False, eps = self.eps)
        scale4b13_branch2b = bn4b13_branch2b
        res4b13_branch2b_relu = mx.symbol.Activation(name='res4b13_branch2b_relu', data=scale4b13_branch2b,
                                                     act_type='relu')
        res4b13_branch2c = mx.symbol.Convolution(name='res4b13_branch2c', data=res4b13_branch2b_relu, num_filter=1024,
                                                 pad=(0, 0), kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b13_branch2c = mx.symbol.BatchNorm(name='bn4b13_branch2c', data=res4b13_branch2c, use_global_stats=True,
                                              fix_gamma=False, eps = self.eps)
        scale4b13_branch2c = bn4b13_branch2c
        res4b13 = mx.symbol.broadcast_add(name='res4b13', *[res4b12_relu, scale4b13_branch2c])
        res4b13_relu = mx.symbol.Activation(name='res4b13_relu', data=res4b13, act_type='relu')
        res4b14_branch2a = mx.symbol.Convolution(name='res4b14_branch2a', data=res4b13_relu, num_filter=256, pad=(0, 0),
                                                 kernel=(1, 1), stride=(1, 1), no_bias=True)
        bn4b14_branch2a = mx.symbol.BatchNorm(name='bn4b14_branch2a', data=res4b14_branch2a, use_global_stats=True,
                                              fix_gamma=False, eps = self.eps)
        scale4b14_branch2a = bn4b14_branch2a