def __init__(self):
super(WholeNetwork, self).__init__()
self.vgg16_extractor = VGG16Extractor()
self.lm_branch = const.LM_BRANCH(const.LM_SELECT_VGG_CHANNEL)
self.downsample = nn.Upsample((28, 28),
mode='bilinear',
align_corners=False)
self.attention_pred_net = CustomUnetGenerator(512 + 1,
512,
num_downs=2,
ngf=32,
last_act='tanh')
self.pooled_4 = nn.MaxPool2d(kernel_size=2, stride=2)
self.conv5_1 = nn.Conv2d(512, 512, 3, padding=1)
self.conv5_2 = nn.Conv2d(512, 512, 3, padding=1)
self.conv5_3 = nn.Conv2d(512, 512, 3, padding=1)
conv5_para_vgg16 = [
self.vgg16_extractor.vgg[-7].state_dict(),
self.vgg16_extractor.vgg[-5].state_dict(),
self.vgg16_extractor.vgg[-3].state_dict(),
]
self.conv5_1.load_state_dict(conv5_para_vgg16[0])
self.conv5_2.load_state_dict(conv5_para_vgg16[1])
self.conv5_3.load_state_dict(conv5_para_vgg16[2])
self.pooled_5 = nn.MaxPool2d(kernel_size=2, stride=2)
self.category_fc1 = nn.Linear(512 * 7 * 7, 1024)
self.category_fc2 = nn.Linear(1024, 48)
self.attr_fc1 = nn.Linear(512 * 7 * 7, 1024)
self.attr_fc2 = nn.Linear(1024, 1000 * 2)
self.category_loss_func = torch.nn.CrossEntropyLoss()
self.attr_loss_func = torch.nn.CrossEntropyLoss(weight=torch.tensor(
[const.WEIGHT_ATTR_NEG, const.WEIGHT_ATTR_POS]).to(const.device))
def __init__(self, input_size, output_size, hidden_dim, n_layers):
super(SimpleModel, self).__init__()
self.vgg16_extractor = VGG16Extractor()
# Defining some parameters
self.hidden_dim = hidden_dim
self.n_layers = n_layers
#Defining the layers
# RNN Layer
self.rnn = nn.RNN(input_size, hidden_dim, n_layers, batch_first=True)
# Fully connected layer
self.fc = nn.Linear(hidden_dim, output_size)
def __init__(self, use_iorn=False):
super(LandmarkExpNetwork, self).__init__()
self.vgg16_extractor = VGG16Extractor(use_iorn=use_iorn)
self.lm_branch = const.LM_BRANCH(const.LM_SELECT_VGG_CHANNEL)