def __init__(self,
vgg_name,
nh=1,
n_h=512,
dropout_prob=0.25,
sm_type='softmax'):
super(VGG, self).__init__()
self.dropout_prob = dropout_prob
self.n_hidden = nh
self.hidden_size = n_h
self.sm_type = sm_type
self.features = self._make_layers(cfg[vgg_name])
self.avgpool = nn.AdaptiveAvgPool2d((7, 7))
self.lin_proj = nn.Sequential(nn.Linear(512 * 7 * 7, 4096),
nn.ReLU(True), nn.Dropout(),
nn.Linear(4096, 512))
if sm_type == 'softmax':
self.out_proj = Softmax(input_features=512, output_features=64)
elif sm_type == 'am_softmax':
self.out_proj = AMSoftmax(input_features=512, output_features=64)
else:
raise NotImplementedError
self.bin_classifier = self.make_bin_layers(n_in=2 * 512,
n_h_layers=nh,
h_size=n_h,
dropout_p=dropout_prob)
def __init__(self, block, num_blocks, sm_type, num_classes=64):
super(ResNet, self).__init__()
self.in_planes = 64
self.conv1 = nn.Conv2d(3,
64,
kernel_size=3,
stride=1,
padding=1,
bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, num_blocks[0], stride=1)
self.layer2 = self._make_layer(block, 128, num_blocks[1], stride=2)
self.layer3 = self._make_layer(block, 256, num_blocks[2], stride=2)
self.layer4 = self._make_layer(block, 512, num_blocks[3], stride=2)
if sm_type == 'softmax':
self.out_proj = Softmax(input_features=512 * block.expansion,
output_features=num_classes)
elif sm_type == 'am_softmax':
self.out_proj = AMSoftmax(input_features=512 * block.expansion,
output_features=num_classes)
else:
raise NotImplementedError
def __init__(self,
vgg_name,
nh=1,
n_h=512,
dropout_prob=0.25,
sm_type='softmax'):
super(VGG, self).__init__()
self.dropout_prob = dropout_prob
self.n_hidden = nh
self.hidden_size = n_h
self.sm_type = sm_type
self.features = self._make_layers(cfg[vgg_name])
if sm_type == 'softmax':
self.out_proj = Softmax(input_features=512, output_features=10)
elif sm_type == 'am_softmax':
self.out_proj = AMSoftmax(input_features=512, output_features=10)
else:
raise NotImplementedError
self.bin_classifier = self.make_bin_layers(n_in=2 * 512,
n_h_layers=nh,
h_size=n_h,
dropout_p=dropout_prob)
def __init__(self, block, num_blocks, nh, n_h, sm_type, num_classes=64, dropout_prob=0.25): super(ResNet, self).__init__() self.dropout_prob = dropout_prob self.n_hidden = nh self.hidden_size = n_h self.sm_type = sm_type self.in_planes = 64 self.conv1 = nn.Conv2d(3, 64, kernel_size=5, stride=1, padding=1, bias=False) self.bn1 = nn.BatchNorm2d(64) self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) self.layer1 = self._make_layer(block, 64, num_blocks[0], stride=1) self.layer2 = self._make_layer(block, 128, num_blocks[1], stride=2) self.layer3 = self._make_layer(block, 256, num_blocks[2], stride=2) self.layer4 = self._make_layer(block, 512, num_blocks[3], stride=2) if sm_type=='softmax': self.out_proj=Softmax(input_features=512*block.expansion, output_features=num_classes) elif sm_type=='am_softmax': self.out_proj=AMSoftmax(input_features=512*block.expansion, output_features=num_classes) else: raise NotImplementedError self.classifier = self.make_bin_layers(n_in=2*512*block.expansion, n_h_layers=nh, h_size=n_h, dropout_p=dropout_prob)
def __init__(self,
block,
nblocks,
nh,
n_h,
sm_type,
growth_rate=12,
reduction=0.5,
num_classes=10,
dropout_prob=0.25):
super(DenseNet, self).__init__()
self.growth_rate = growth_rate
num_planes = 2 * growth_rate
self.conv1 = nn.Conv2d(3,
num_planes,
kernel_size=3,
padding=1,
bias=False)
self.dense1 = self._make_dense_layers(block, num_planes, nblocks[0])
num_planes += nblocks[0] * growth_rate
out_planes = int(math.floor(num_planes * reduction))
self.trans1 = Transition(num_planes, out_planes)
num_planes = out_planes
self.dense2 = self._make_dense_layers(block, num_planes, nblocks[1])
num_planes += nblocks[1] * growth_rate
out_planes = int(math.floor(num_planes * reduction))
self.trans2 = Transition(num_planes, out_planes)
num_planes = out_planes
self.dense3 = self._make_dense_layers(block, num_planes, nblocks[2])
num_planes += nblocks[2] * growth_rate
out_planes = int(math.floor(num_planes * reduction))
self.trans3 = Transition(num_planes, out_planes)
num_planes = out_planes
self.dense4 = self._make_dense_layers(block, num_planes, nblocks[3])
num_planes += nblocks[3] * growth_rate
self.bn = nn.BatchNorm2d(num_planes)
if sm_type == 'softmax':
self.out_proj = Softmax(input_features=num_planes,
output_features=num_classes)
elif sm_type == 'am_softmax':
self.out_proj = AMSoftmax(input_features=num_planes,
output_features=num_classes)
else:
raise NotImplementedError
self.classifier = self.make_bin_layers(n_in=2 * num_planes,
n_h_layers=nh,
h_size=n_h,
dropout_p=dropout_prob)
def __init__(self, vgg_name, sm_type='softmax'):
super(VGG, self).__init__()
self.features = self._make_layers(cfg[vgg_name])
if sm_type == 'softmax':
self.out_proj = Softmax(input_features=512, output_features=10)
elif sm_type == 'am_softmax':
self.out_proj = AMSoftmax(input_features=512, output_features=10)
else:
raise NotImplementedError
def __init__(self, block, layers, sm_type, num_classes=1000, zero_init_residual=False,
groups=1, width_per_group=64, replace_stride_with_dilation=None,
norm_layer=None):
super(ResNet, self).__init__()
if norm_layer is None:
norm_layer = nn.BatchNorm2d
self._norm_layer = norm_layer
self.n_classes = num_classes
self.inplanes = 64
self.dilation = 1
if replace_stride_with_dilation is None:
# each element in the tuple indicates if we should replace
# the 2x2 stride with a dilated convolution instead
replace_stride_with_dilation = [False, False, False]
if len(replace_stride_with_dilation) != 3:
raise ValueError("replace_stride_with_dilation should be None "
"or a 3-element tuple, got {}".format(replace_stride_with_dilation))
self.groups = groups
self.base_width = width_per_group
self.conv1 = nn.Conv2d(3, self.inplanes, kernel_size=7, stride=2, padding=3, bias=False)
self.bn1 = norm_layer(self.inplanes)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2, dilate=replace_stride_with_dilation[0])
self.layer3 = self._make_layer(block, 256, layers[2], stride=2, dilate=replace_stride_with_dilation[1])
self.layer4 = self._make_layer(block, 512, layers[3], stride=2, dilate=replace_stride_with_dilation[2])
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.lin_proj = nn.Sequential(nn.Linear(512*block.expansion, 128))
if sm_type=='softmax':
self.out_proj=Softmax(input_features=512*block.expansion, output_features=num_classes)
elif sm_type=='am_softmax':
self.out_proj=AMSoftmax(input_features=512*block.expansion, output_features=num_classes)
else:
raise NotImplementedError
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
# Zero-initialize the last BN in each residual branch,
# so that the residual branch starts with zeros, and each residual block behaves like an identity.
# This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
if zero_init_residual:
for m in self.modules():
if isinstance(m, Bottleneck):
nn.init.constant_(m.bn3.weight, 0)
elif isinstance(m, BasicBlock):
nn.init.constant_(m.bn2.weight, 0)
def __init__(self,
block,
nblocks,
sm_type,
growth_rate=12,
reduction=0.5,
num_classes=1000):
super(DenseNet, self).__init__()
self.growth_rate = growth_rate
self.n_classes = num_classes
num_planes = 2 * growth_rate
self.conv1 = nn.Conv2d(3,
num_planes,
kernel_size=7,
padding=1,
bias=False)
self.dense1 = self._make_dense_layers(block, num_planes, nblocks[0])
num_planes += nblocks[0] * growth_rate
out_planes = int(math.floor(num_planes * reduction))
self.trans1 = Transition(num_planes, out_planes)
num_planes = out_planes
self.dense2 = self._make_dense_layers(block, num_planes, nblocks[1])
num_planes += nblocks[1] * growth_rate
out_planes = int(math.floor(num_planes * reduction))
self.trans2 = Transition(num_planes, out_planes)
num_planes = out_planes
self.dense3 = self._make_dense_layers(block, num_planes, nblocks[2])
num_planes += nblocks[2] * growth_rate
out_planes = int(math.floor(num_planes * reduction))
self.trans3 = Transition(num_planes, out_planes)
num_planes = out_planes
self.dense4 = self._make_dense_layers(block, num_planes, nblocks[3])
num_planes += nblocks[3] * growth_rate
self.bn = nn.BatchNorm2d(num_planes)
self.lin_proj = nn.Sequential(nn.Linear(1024 * 6 * 6, 1024),
nn.ReLU(True), nn.Dropout(0.1),
nn.Linear(1024, 512))
if sm_type == 'softmax':
self.out_proj = Softmax(input_features=512,
output_features=num_classes)
elif sm_type == 'am_softmax':
self.out_proj = AMSoftmax(input_features=512,
output_features=num_classes)
else:
raise NotImplementedError
def __init__(self, vgg_name, sm_type='softmax'):
super(VGG, self).__init__()
self.features = self._make_layers(cfg[vgg_name])
self.avgpool = nn.AdaptiveAvgPool2d((7, 7))
self.lin_proj = nn.Sequential(nn.Linear(512 * 7 * 7, 4096),
nn.ReLU(True), nn.Dropout(),
nn.Linear(4096, 512))
if sm_type == 'softmax':
self.out_proj = Softmax(input_features=512, output_features=600)
elif sm_type == 'am_softmax':
self.out_proj = AMSoftmax(input_features=512, output_features=600)
else:
raise NotImplementedError
def __init__(self, vgg_name, sm_type='softmax', n_classes=1000):
super(VGG, self).__init__()
self.n_classes = n_classes
self.features = self._make_layers(cfg[vgg_name])
self.avgpool = nn.AdaptiveAvgPool2d((7, 7))
self.lin_proj = nn.Sequential(nn.Linear(512 * 7 * 7, 128))
if sm_type == 'softmax':
self.out_proj = Softmax(input_features=512 * 7 * 7,
output_features=self.n_classes)
elif sm_type == 'am_softmax':
self.out_proj = AMSoftmax(input_features=512 * 7 * 7,
output_features=self.n_classes)
else:
raise NotImplementedError
def __init__(self, vgg_name, nh=1, n_h=512, emb_size=128, dropout_prob=0.25, sm_type='softmax', n_classes=1000, r_proj_size=0): super(VGG, self).__init__() self.dropout_prob = dropout_prob self.n_hidden = nh self.hidden_size = n_h self.sm_type = sm_type self.n_classes = n_classes self.emb_size = emb_size self.r_proj_size = r_proj_size self.features = self._make_layers(cfg[vgg_name]) self.avgpool = nn.AdaptiveAvgPool2d((7, 7)) self.lin_proj = nn.Sequential(nn.Linear(512 * 7 * 7, self.emb_size)) if sm_type=='softmax': self.out_proj=Softmax(input_features=self.emb_size, output_features=self.n_classes) elif sm_type=='am_softmax': self.out_proj=AMSoftmax(input_features=self.emb_size, output_features=self.n_classes) else: raise NotImplementedError self.bin_classifier = self.make_bin_layers(n_in=2*self.emb_size, n_h_layers=nh, h_size=n_h, dropout_p=dropout_prob)