def __init__(self, base_model_architecture="resnet50_128", num_clusters=8, vset_dim=128,
pooling='vlad', vlad_v2=False):
super(HashSetNet, self).__init__()
if base_model_architecture == "resnet50_128":
self.base_model = resnet50_128(ROOT_DIR + '/Models_weights/resnet50_128.pth')
dim = 128
elif base_model_architecture == "senet50_128":
self.base_model = senet50_128(ROOT_DIR + '/Models_weights/senet50_128.pth')
dim = 128
elif base_model_architecture == "resnet50_2048":
self.base_model = resnet50_ft(ROOT_DIR + '/Models_weights/resnet50_ft_dims_2048.pth')
dim = 2048
elif base_model_architecture == "senet50_2048":
self.base_model = senet50_ft(ROOT_DIR + '/Models_weights/senet50_ft_dims_2048.pth')
dim = 2048
self.pooling = pooling
if self.pooling == 'vlad':
self.net_vlad = NetVLAD(num_clusters=num_clusters, dim=dim, vset_dim=vset_dim,
vlad_v2=vlad_v2, normalize_input=True)
elif self.pooling == 'gem':
self.gem_pooling = GeM(p=3, eps=1e-6)
elif self.pooling == 'sum':
self.sum_pooling = SumPooling()
self.bn_x = nn.BatchNorm1d(dim, affine=False)
def __init__(self, model_type="resnet50_128"):
super(Net, self).__init__()
if model_type == "resnet50_128":
self.base_model = resnet50_128(ROOT_DIR + '/Models_weights/resnet50_128.pth')
self.encoder_dim = 128
elif model_type == "senet50_128":
self.base_model = senet50_128(ROOT_DIR + '/Models_weights/senet50_128.pth')
self.encoder_dim = 128
elif model_type == "resnet50_2048":
self.base_model = resnet50_ft(ROOT_DIR + '/Models_weights/resnet50_ft_dims_2048.pth')
self.encoder_dim = 2048
elif model_type == "senet50_2048":
self.base_model = senet50_ft(ROOT_DIR + '/Models_weights/senet50_ft_dims_2048.pth')
self.encoder_dim = 2048
def __init__(self,
base_model_architecture="resnet50_128",
num_clusters=8,
vset_dim=128,
pooling='vlad',
vlad_v2=False,
step_size=200):
super(HashSetNet, self).__init__()
if base_model_architecture == "resnet50_128":
self.base_model = resnet50_128(ROOT_DIR +
'/Models_weights/resnet50_128.pth')
dim = 128
elif base_model_architecture == "senet50_128":
self.base_model = senet50_128(ROOT_DIR +
'/Models_weights/senet50_128.pth')
dim = 128
elif base_model_architecture == "resnet50_2048":
self.base_model = resnet50_ft(
ROOT_DIR + '/Models_weights/resnet50_ft_dims_2048.pth')
dim = 2048
elif base_model_architecture == "senet50_2048":
self.base_model = senet50_ft(
ROOT_DIR + '/Models_weights/senet50_ft_dims_2048.pth')
dim = 2048
self.pooling = pooling
if self.pooling == 'vlad':
self.net_vlad = NetVLAD(num_clusters=num_clusters,
dim=dim,
vset_dim=vset_dim,
vlad_v2=vlad_v2,
normalize_input=True)
elif self.pooling == 'gem':
self.gem_pooling = GeM(p=3, eps=1e-6)
elif self.pooling == 'sum':
self.sum_pooling = SumPooling()
self.bn_x = nn.BatchNorm1d(dim, affine=False)
self.iter_num = 0
self.step_size = step_size
self.gamma = 0.005
self.power = 0.5
self.init_scale = 1.0
self.activation = nn.Tanh()
self.scale = self.init_scale
self.first_flag = True
def __init__(self, base_model_architecture="resnet50_128", num_clusters=8, vlad_dim=128, vlad_v2=False):
super(SetNet, self).__init__()
if base_model_architecture == "resnet50_128":
self.base_model = resnet50_128(ROOT_DIR + '/Models_weights/resnet50_128.pth')
dim = 128
elif base_model_architecture == "senet50_128":
self.base_model = senet50_128(ROOT_DIR + '/Models_weights/senet50_128.pth')
dim = 128
elif base_model_architecture == "resnet50_2048":
self.base_model = resnet50_ft(ROOT_DIR + '/Models_weights/resnet50_ft_dims_2048.pth')
dim = 2048
elif base_model_architecture == "senet50_2048":
self.base_model = senet50_ft(ROOT_DIR + '/Models_weights/senet50_ft_dims_2048.pth')
dim = 2048
self.net_vlad = NetVLAD(num_clusters=num_clusters, dim=dim, vlad_dim=vlad_dim,
vlad_v2=vlad_v2, normalize_input=True)
self.bn_x = nn.BatchNorm1d(dim, affine=False)
def __init__(self, base_model_architecture="resnet50_128"):
super(Baseline, self).__init__()
if base_model_architecture == "resnet50_128":
self.base_model = resnet50_128(ROOT_DIR +
'/Models_weights/resnet50_128.pth')
dim = 128
elif base_model_architecture == "senet50_128":
self.base_model = senet50_128(ROOT_DIR +
'/Models_weights/senet50_128.pth')
dim = 128
elif base_model_architecture == "resnet50_2048":
self.base_model = resnet50_ft(
ROOT_DIR + '/Models_weights/resnet50_ft_dims_2048.pth')
dim = 2048
elif base_model_architecture == "senet50_2048":
self.base_model = senet50_ft(
ROOT_DIR + '/Models_weights/senet50_ft_dims_2048.pth')
dim = 2048
self.sum_pooling = SumPooling()
self.bn_x = nn.BatchNorm1d(dim, affine=False)