def __init__(self,
class_num=31,
alpha=1.0,
beta=1.0,
gamma=1.0,
use_gpu=False,
writer=None):
self.use_gpu = use_gpu
self.alpha = alpha
self.beta = beta
self.gamma = gamma
self.writer = writer
self.f_net = FeatureExtractor()
high_feature_dim = self.f_net.output_dim()
self.bn = BatchNorm1d(high_feature_dim)
self.c_net_a = FeatureClassifier(feature_dim=high_feature_dim,
class_num=class_num)
self.c_net_b = FeatureClassifier(feature_dim=high_feature_dim,
class_num=class_num)
if self.use_gpu:
self.f_net = self.f_net.cuda()
self.c_net_a = self.c_net_a.cuda()
self.c_net_b = self.c_net_b.cuda()
self.bn = self.bn.cuda()
def __init__(self, input_dim, virtual_batch_size=64, momentum=0.01):
super(GhostNormalization, self).__init__()
self.input_dim = input_dim
self.virtual_batch_size = virtual_batch_size
self.bn = BatchNorm1d(self.input_dim, momentum=momentum, affine=False)
self.weight = Parameter(torch.Tensor(self.input_dim))
self.bias = Parameter(torch.Tensor(self.input_dim))
init.uniform_(self.bias)
init.uniform_(self.weight)
def __init__(self,
channels_in=1,
channels_out=1,
kernel_size=1,
use_batch_norm=True,
apply_relu=True):
super(Block, self).__init__()
self.conv = nn.Conv1d(channels_in, channels_out, kernel_size)
self.relu = nn.ReLU()
self.batchnorm = BatchNorm1d(channels_out)
self.use_batch_norm = use_batch_norm
self.apply_relu = apply_relu
def __init__(self, encoder_hidden, decoder_hidden, output_dim,
embedding_dim, sos_token, device):
super().__init__()
self.decoder_hidden = decoder_hidden
self.gru = nn.GRU(2 * encoder_hidden + embedding_dim, decoder_hidden)
self.attn = AttentionNetwork(encoder_hidden, decoder_hidden)
self.embedding = nn.Embedding(output_dim, embedding_dim)
self.output_decoder = nn.Linear(decoder_hidden, output_dim)
self.batch_norm = BatchNorm1d(output_dim)
self.output_dim = output_dim
self.device = device
self.sos_token = sos_token
def __init__(self, feature_dim=256, class_num=31):
super(FeatureClassifier, self).__init__()
self.classifier_layer_1 = nn.Linear(feature_dim, feature_dim // 2)
self.classifier_layer_2 = nn.Linear(feature_dim // 2, class_num)
self.bn = BatchNorm1d(feature_dim // 2)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(0.5)
self.softmax = nn.Softmax()
## initialization
self.classifier_layer_1.weight.data.normal_(0, 0.01)
self.classifier_layer_1.bias.data.fill_(0.0)
self.classifier_layer_2.weight.data.normal_(0, 0.01)
self.classifier_layer_2.bias.data.fill_(0.0)
def __init__(self):
super(FeatureExtractor, self).__init__()
## set base network
model_resnet50 = models.resnet50(pretrained=True)
self.conv1 = model_resnet50.conv1
self.bn1 = model_resnet50.bn1
self.relu = model_resnet50.relu
self.maxpool = model_resnet50.maxpool
self.layer1 = model_resnet50.layer1
self.layer2 = model_resnet50.layer2
self.layer3 = model_resnet50.layer3
self.layer4 = model_resnet50.layer4
self.avgpool = model_resnet50.avgpool
self.high_dim = model_resnet50.fc.in_features
self.bn = BatchNorm1d(self.high_dim)
def __init__(self,
class_num=31,
trade_off=1.0,
use_gpu=False,
writer=None):
self.use_gpu = use_gpu
self.trade_off = trade_off
self.writer = writer
self.f_net = FeatureExtractor()
high_feature_dim = self.f_net.output_dim()
self.bn = BatchNorm1d(high_feature_dim, affine=False).cuda()
#feature_dim = adaptor_dim
self.c_net_a = FeatureClassifier(feature_dim=high_feature_dim,
class_num=class_num)
self.c_net_b = FeatureClassifier(feature_dim=high_feature_dim,
class_num=class_num)
if self.use_gpu:
self.f_net = self.f_net.cuda()
self.c_net_a = self.c_net_a.cuda()
self.c_net_b = self.c_net_b.cuda()
def __init__(self, feature_dim=2048, output_dim=256, final_bn=True):
super(SingleLayerFeatureExtractor, self).__init__()
self.classifier_layer_1 = nn.Linear(feature_dim, output_dim)
self.classifier_layer_1.weight.data.normal_(0, 0.005)
self.classifier_layer_1.bias.data.fill_(0.1)
self.bn = BatchNorm1d(output_dim)