def __init__(self, in_channel_block, in_channel_clf, num_classes, cls,
hidden_channel, attention_1, attention_2):
super(ClassifierModuleMiddle, self).__init__()
self.relu = nn.ReLU(inplace=False)
self.BN = nn.BatchNorm2d(in_channel_block)
self.linear_h = nn.Linear(in_channel_block + in_channel_clf,
hidden_channel)
self.linear = nn.Linear(hidden_channel, num_classes)
if attention_1 == 'se':
self.attention_1 = SELayer(in_channel_block, 16)
elif attention_1 == 'scan':
self.attention_1 = ScanLayer(in_channel_block)
elif attention_1 == 'no':
self.attention_1 = lambda x: x # dummy
else:
raise ValueError
self.cls = cls
if self.cls != 0:
self.b0 = nn.ParameterList(
[nn.Parameter(torch.zeros(1, hidden_channel))])
self.linear_bw = nn.Linear(hidden_channel,
in_channel_block + in_channel_clf)
self.BN1d = nn.BatchNorm1d(hidden_channel)
if attention_2 == 'yes':
self.attention_2 = LinearLayer(in_channel_clf)
elif attention_2 == 'no':
self.attention_2 = lambda x: x # dummy
else:
raise ValueError
def __init__(self,
inplanes,
planes,
stride=1,
downsample=None,
use_att=False,
att_mode='ours'):
super(Bottleneck, self).__init__()
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
self.bn1 = nn.BatchNorm2d(planes)
self.conv2 = nn.Conv2d(planes,
planes,
kernel_size=3,
stride=stride,
padding=1,
bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.conv3 = nn.Conv2d(planes,
planes * self.expansion,
kernel_size=1,
bias=False)
self.bn3 = nn.BatchNorm2d(planes * self.expansion)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
self.use_att = use_att
if use_att:
assert att_mode in ['ours', 'se']
if att_mode == 'ours':
self.att = AttentionModule(planes * self.expansion)
elif att_mode == 'se':
self.att = SELayer(planes * self.expansion)
def __init__(self, in_channel_block, num_classes, hidden_channel):
super(ClassifierModuleFirst, self).__init__()
self.relu = nn.ReLU(inplace=False)
self.BN = nn.BatchNorm2d(in_channel_block)
self.se = SELayer(in_channel_block, 16)
self.linear_h = nn.Linear(in_channel_block, hidden_channel)
self.linear = nn.Linear(hidden_channel, num_classes)
self.BN1d = nn.BatchNorm1d(hidden_channel)
def __init__(self, in_channel_block, num_classes, hidden_channel,
attention):
super(ClassifierModuleFirst, self).__init__()
self.relu = nn.ReLU(inplace=False)
self.BN = nn.BatchNorm2d(in_channel_block)
if attention == 'se':
self.attention = SELayer(in_channel_block, 16)
elif attention == 'scan':
self.attention = ScanLayer(in_channel_block)
elif attention == 'no':
self.attention = lambda x: x # dummy
else:
raise ValueError
self.linear_h = nn.Linear(in_channel_block, hidden_channel)
self.linear = nn.Linear(hidden_channel, num_classes)
self.BN1d = nn.BatchNorm1d(hidden_channel)
def __init__(self,
inplanes,
planes,
stride=1,
downsample=None,
use_att=False,
att_mode='ours'):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes, stride)
self.bn1 = nn.BatchNorm2d(planes)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
self.bn2 = nn.BatchNorm2d(planes)
self.downsample = downsample
self.stride = stride
self.use_att = use_att
if use_att:
assert att_mode in ['ours', 'se']
if att_mode == 'ours':
self.att = AttentionModule(planes)
elif att_mode == 'se':
self.att = SELayer(planes)
def __init__(self,
in_channel_block,
in_channel_clf,
num_classes,
cls,
hidden_channel,
score_layer=False):
super(ClassifierModuleMiddle, self).__init__()
self.relu = nn.ReLU(inplace=False)
self.BN = nn.BatchNorm2d(in_channel_block)
self.se = SELayer(in_channel_block, 16)
self.linear_h = nn.Linear(in_channel_block + in_channel_clf,
hidden_channel)
self.linear = nn.Linear(hidden_channel, num_classes)
self.cls = cls
if self.cls != 0:
self.b0 = nn.ParameterList(
[nn.Parameter(torch.zeros(1, hidden_channel))])
self.linear_bw = nn.Linear(hidden_channel,
in_channel_block + in_channel_clf)
self.BN1d = nn.BatchNorm1d(hidden_channel)