def __init__(self,
dataset,
decoder,
encoder=None,
encoder_latent=None,
p0_z=None,
device=None):
super().__init__()
if p0_z is None:
p0_z = dist.Normal(torch.tensor([]), torch.tensor([]))
self.decoder = decoder.to(device)
if encoder_latent is not None:
self.encoder_latent = encoder_latent.to(device)
else:
self.encoder_latent = None
if encoder is not None:
self.encoder = encoder.to(device)
else:
self.encoder = None
# init category_center
category_count = len(dataset.metadata)
self.category_count = category_count
self.category_centers = torch.zeros(category_count,
decoder.c_dim).to(device)
self.pre_category_centers = torch.rand(category_count,
decoder.c_dim).to(device)
self._device = device
self.p0_z = p0_z
def __init__(self, cfg, decoder, encoder=None, dann_discriminator=None, encoder_latent=None, p0_z=None,
device=None):
super().__init__()
if p0_z is None:
p0_z = dist.Normal(torch.tensor([]), torch.tensor([]))
self.decoder = decoder.to(device)
if encoder_latent is not None:
self.encoder_latent = encoder_latent.to(device)
else:
self.encoder_latent = None
if encoder is not None:
self.encoder = encoder.to(device)
else:
self.encoder = None
if dann_discriminator is not None:
self.dann_discriminator = dann_discriminator.to(device)
else:
self.dann_discriminator = None
self._device = device
self.p0_z = p0_z
def __init__(
self,
dataset,
decoder1,
decoder2,
decoder3,
encoder=None,
encoder_latent=None,
p0_z=None,
device=None,
use_local_feature=False,
logits2_ratio=1.,
logits1_ratio=1.,
local_feature_mask=False,
):
super().__init__()
if p0_z is None:
p0_z = dist.Normal(torch.tensor([]), torch.tensor([]))
self.decoder1 = decoder1.to(device)
self.decoder2 = decoder2.to(device)
self.decoder3 = decoder3.to(device)
self.use_local_feature = use_local_feature
self.local_feature_mask = local_feature_mask
if self.local_feature_mask:
print('using local feature predict mask')
if encoder_latent is not None:
self.encoder_latent = encoder_latent.to(device)
else:
self.encoder_latent = None
if encoder is not None:
self.encoder = encoder.to(device)
else:
self.encoder = None
# init category_center
category_count = len(dataset.metadata)
self.category_count = category_count
self._device = device
self.p0_z = p0_z
self.logits2_ratio = logits2_ratio
self.logits1_ratio = logits1_ratio