def __init__(self, block, layers, channels, classes=-1, thumbnail=False,
embedding_size=512, weight_norm=False, feature_norm=False,
need_cls_layer=True, **kwargs):
super(SE_ResNetV2, self).__init__(classes, embedding_size, weight_norm,
feature_norm, need_cls_layer, **kwargs)
assert len(layers) == len(channels) - 1
with self.name_scope():
self.features = nn.HybridSequential(prefix='')
self.features.add(nn.BatchNorm(scale=False, center=False))
if thumbnail:
self.features.add(_conv3x3(channels[0], 1, 0))
else:
self.features.add(nn.Conv2D(channels[0], 3, 1, 1, use_bias=False))
self.features.add(nn.BatchNorm())
self.features.add(nn.PReLU())
self.features.add(nn.MaxPool2D(3, 2, 1))
in_channels = channels[0]
for i, num_layer in enumerate(layers):
stride = 1 if i == 0 else 2
self.features.add(self._make_layer(block, num_layer, channels[i + 1],
stride, i + 1, in_channels=in_channels))
in_channels = channels[i + 1]
self.features.add(nn.BatchNorm())
self.features.add(nn.PReLU())
self.features.add(nn.GlobalAvgPool2D())
self.features.add(nn.Flatten())
self.features.add(nn.Dense(embedding_size, use_bias=False))
self.features.add(nn.BatchNorm(scale=False, center=False))
self.features.add(nn.PReLU())
def __init__(self,
channels,
stride,
downsample=False,
in_channels=0,
**kwargs):
super(SE_BottleneckV2, self).__init__(**kwargs)
self.bn1 = nn.BatchNorm()
self.prelu1 = nn.PReLU()
self.conv1 = nn.Conv2D(channels // 4,
kernel_size=1,
strides=1,
use_bias=False)
self.bn2 = nn.BatchNorm()
self.prelu2 = nn.PReLU()
self.conv2 = _conv3x3(channels // 4, stride, channels // 4)
self.bn3 = nn.BatchNorm()
self.prelu3 = nn.PReLU()
self.conv3 = nn.Conv2D(channels,
kernel_size=1,
strides=1,
use_bias=False)
if downsample:
self.downsample = nn.Conv2D(channels,
1,
stride,
use_bias=False,
in_channels=in_channels)
else:
self.downsample = None
self.se = SELayer(channels, channels)