def __init__()

in utils/gluon/utils/resnetv1.py [0:0]

• 33 lines of code
• 8 McCabe index (conditional complexity)

    def __init__(self, in_planes, mid_planes, out_planes, strides=1,
                 norm_kwargs=None, last_gamma=False, name_prefix=None,
                 kernel_sizes=(1,3,1), **kwargs):
        super(_bL_Stem, self).__init__(prefix=name_prefix)
        with self.name_scope():
            # extract information
            self.conv1 = gluon.nn.Conv2D(channels=mid_planes, in_channels=in_planes,
                                  kernel_size=kernel_sizes[0],
			          padding=int((kernel_sizes[0]-1)/2),
                                  use_bias=False, prefix='conv1_')
            self.bn1 = gluon.nn.BatchNorm(in_channels=mid_planes, prefix='bn1_',
                                  **({} if norm_kwargs is None else norm_kwargs))
            self.relu1 = gluon.nn.Activation('relu')
            # capture spatial relations
            self.conv2 = gluon.nn.Conv2D(channels=mid_planes, in_channels=mid_planes,
                                  kernel_size=kernel_sizes[1],
                                  padding=int((kernel_sizes[1]-1)/2),
                                  strides=strides, use_bias=False, prefix='conv2_')
            self.bn2 = gluon.nn.BatchNorm(in_channels=mid_planes, prefix='bn2_',
                                  **({} if norm_kwargs is None else norm_kwargs))
            self.relu2 = gluon.nn.Activation('relu')
            # embeding back to information highway
            self.conv3 = gluon.nn.Conv2D(channels=out_planes, in_channels=mid_planes,
                                  kernel_size=kernel_sizes[2],
                                  padding=int((kernel_sizes[2]-1)/2),
                                  use_bias=False, prefix='conv3_')
            self.bn3 = gluon.nn.BatchNorm(in_channels=out_planes, prefix='bn3_',
                                  gamma_initializer='zeros' if last_gamma else 'ones',
                                  **({} if norm_kwargs is None else norm_kwargs))

            # this relue is added after fusion
            self.relu3 = gluon.nn.Activation('relu')

            if strides != 1 or in_planes != out_planes:
                self.conv4 = gluon.nn.Conv2D(channels=out_planes, in_channels=in_planes,
                                  kernel_size=1, strides=strides, use_bias=False,
                                  prefix='conv4_')
                self.bn4 = gluon.nn.BatchNorm(in_channels=out_planes, prefix='bn4_',
                                  **({} if norm_kwargs is None else norm_kwargs))