def __init__(self, n_layer, n_class=None, initialW=None, fc_kwargs={}):
blocks = self._blocks[n_layer]
if initialW is None:
initialW = initializers.HeNormal(scale=1., fan_option='fan_out')
if 'initialW' not in fc_kwargs:
fc_kwargs['initialW'] = initializers.Normal(scale=0.01)
kwargs = {
'initialW': initialW,
'stride_first': True,
'add_block': 'gc'
}
super(GCResNet, self).__init__()
with self.init_scope():
self.conv1 = Conv2DBNActiv(None,
64,
7,
2,
3,
nobias=True,
initialW=initialW)
self.pool1 = lambda x: F.max_pooling_2d(x, ksize=3, stride=2)
self.res2 = ResBlock(blocks[0], None, 64, 256, 1, **kwargs)
self.res3 = ResBlock(blocks[1], None, 128, 512, 2, **kwargs)
self.res4 = ResBlock(blocks[2], None, 256, 1024, 2, **kwargs)
self.res5 = ResBlock(blocks[3], None, 512, 2048, 2, **kwargs)
self.pool5 = lambda x: F.average(x, axis=(2, 3))
self.fc6 = L.Linear(None, n_class, **fc_kwargs)
def __init__(self, n_layer,
n_class=None, scale=4,
pretrained_model=None,
mean=None, initialW=None, fc_kwargs={}):
blocks = self._blocks[n_layer]
self.mean = mean
if initialW is None:
initialW = initializers.HeNormal(scale=1., fan_option='fan_out')
if 'initialW' not in fc_kwargs:
fc_kwargs['initialW'] = initializers.Normal(scale=0.01)
kwargs = {
'scale': scale, 'initialW': initialW, 'stride_first': True,
'add_seblock': True}
super(SERes2Net, self).__init__()
with self.init_scope():
self.conv1 = Conv2DBNActiv(None, 64, 3, 1, 1, nobias=True,
initialW=initialW)
self.res2 = ResBlock(blocks[0], None, 64, 256, 2, **kwargs)
self.res3 = ResBlock(blocks[1], None, 128, 512, 1, **kwargs)
self.res4 = ResBlock(blocks[2], None, 256, 1024, 2, **kwargs)
self.res5 = ResBlock(blocks[3], None, 512, 2048, 1, **kwargs)
self.pool5 = lambda x: F.average(x, axis=(2, 3))
self.fc6 = L.Linear(None, n_class, **fc_kwargs)
def __init__(self,
in_channels,
mid_channels,
out_channels,
stride=1,
scale=1,
dilate=1,
groups=1,
initialW=None,
bn_kwargs={},
residual_conv=False,
stride_first=False,
add_block=None,
aa_kwargs={}):
if stride_first:
first_stride = stride
second_stride = 1
else:
first_stride = 1
second_stride = stride
super(Bottleneck, self).__init__()
with self.init_scope():
self.conv1 = Conv2DBNActiv(in_channels,
mid_channels,
1,
first_stride,
0,
nobias=True,
initialW=initialW,
bn_kwargs=bn_kwargs)
if len(aa_kwargs) > 0:
self.conv2 = Conv2DBNActiv(mid_channels,
mid_channels,
3,
1,
1,
1,
1,
1,
nobias=True,
initialW=initialW,
bn_kwargs=bn_kwargs,
aa_kwargs=aa_kwargs)
elif stride > 1:
self.conv2 = Conv2DBNActiv(mid_channels,
mid_channels,
3,
second_stride,
dilate,
1,
dilate,
groups,
nobias=True,
initialW=initialW,
bn_kwargs=bn_kwargs)
else:
self.conv2 = Conv2DBNActiv(mid_channels,
mid_channels,
3,
second_stride,
dilate,
scale,
dilate,
groups,
nobias=True,
initialW=initialW,
bn_kwargs=bn_kwargs)
self.conv3 = Conv2DBNActiv(mid_channels,
out_channels,
1,
1,
0,
nobias=True,
initialW=initialW,
activ=None,
bn_kwargs=bn_kwargs)
if add_block == 'se':
self.se = SEBlock(out_channels)
elif add_block == 'gc':
self.gc = GCBlock(out_channels)
elif add_block is not None:
raise ValueError
if residual_conv:
self.residual_conv = Conv2DBNActiv(in_channels,
out_channels,
1,
stride,
0,
nobias=True,
initialW=initialW,
activ=None,
bn_kwargs=bn_kwargs)