syncbatchnorm0_gamma' has not been initialized yet because initialization was deferred

[ngunauj]

The part of forward..

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.bn2(x)
        x = self.relu(x)

The part of define..

        self.conv1 = nn.Conv2D(
            channels=64,
            kernel_size=3,
            strides=2,
            padding=1,
            in_channels=3,
            use_bias=False)
        self.bn1 = BatchNorm2d(in_channels=64, momentum=BN_MOMENTUM)
        self.conv2 = nn.Conv2D(
            channels=64,
            kernel_size=3,
            strides=2,
            padding=1,
            in_channels=64,
            use_bias=False)
        self.bn2 = BatchNorm2d(in_channels=64, momentum=BN_MOMENTUM)
        self.relu = nn.Activation('relu')

the BatchNorm2d is

from mxnet.gluon.contrib.nn import SyncBatchNorm
BatchNorm2d = functools.partial(SyncBatchNorm, num_devices=4)

when run the code, the error is syncbatchnorm0_gamma' has not been initialized yet because initialization was deferred.
when i change the SyncBatchNorm to BatchNorm is ok....

i use net.initialize(ctx=[mx.cpu(0)]) to init params