def __init__(self, hidden_dim=10): super(conv_Update, self).__init__() self.hidden_dim = hidden_dim dtype = torch.cuda.FloatTensor self.update = ConvGRU(input_dim=hidden_dim, hidden_dim=hidden_dim, kernel_size=(1, 1), num_layers=1, dtype=dtype, batch_first=True, bias=True, return_all_layers=False)
def __init__(self, hidden_dim=10, paths_len=3): super(conv_Update, self).__init__() self.hidden_dim = hidden_dim # detect if CUDA is available or not use_gpu = torch.cuda.is_available() if use_gpu: dtype = torch.cuda.FloatTensor # computation in GPU else: dtype = torch.FloatTensor self.conv_update = ConvGRU(input_dim=paths_len * hidden_dim, hidden_dim=hidden_dim, kernel_size=(1, 1), num_layers=1, dtype=dtype, batch_first=True, bias=True, return_all_layers=False)
def __init__(self, hidden_dim=10, paths_len=3): super(conv_Update, self).__init__() self.hidden_dim = hidden_dim # self.conv_update = nn.Sequential( # nn.Conv2d((paths_len+1) * hidden_dim, 2 * hidden_dim, kernel_size=3, padding=1, stride=1, bias=False), # BatchNorm2d(2 * hidden_dim), nn.ReLU(inplace=False), # nn.Conv2d(2 * hidden_dim, hidden_dim, kernel_size=3, padding=1, stride=1, bias=False), # BatchNorm2d(hidden_dim), nn.ReLU(inplace=False) # ) # detect if CUDA is available or not use_gpu = torch.cuda.is_available() if use_gpu: dtype = torch.cuda.FloatTensor # computation in GPU else: dtype = torch.FloatTensor self.conv_update = ConvGRU(input_dim=paths_len* hidden_dim, hidden_dim=hidden_dim, kernel_size=(1,1), num_layers=1, dtype=dtype, batch_first=True, bias=True, return_all_layers=False)
def __init__(self, in_dim, hidden_dim=10, inputs_num=1): super(conv_Update, self).__init__() self.hidden_dim = hidden_dim self.update = ConvGRU(input_dim=in_dim, hidden_dim=hidden_dim, inputs_num=inputs_num)