def __init__(self,
c_dim=128,
dim=3,
hidden_dim=128,
scatter_type='max',
unet=False,
unet_kwargs=None,
plane_resolution=None,
grid_resolution=None,
plane_type='xz',
padding=0.1,
n_blocks=5,
pos_encoding=False,
n_channels=3,
plane_net='FCPlanenet'):
super().__init__()
self.c_dim = c_dim
self.num_channels = n_channels
if pos_encoding == True:
dim = 60
self.fc_pos = nn.Linear(dim, 2 * hidden_dim)
self.blocks = nn.ModuleList([
ResnetBlockFC(2 * hidden_dim, hidden_dim) for i in range(n_blocks)
])
self.fc_c = nn.Linear(hidden_dim, c_dim)
planenet_hidden_dim = hidden_dim
self.fc_plane_net = FCPlanenet(n_dim=dim, hidden_dim=hidden_dim)
# Create FC layers based on the number of planes
self.plane_params = nn.ModuleList(
[nn.Linear(planenet_hidden_dim, 3) for i in range(n_channels)])
self.plane_params_hdim = nn.ModuleList(
[nn.Linear(3, hidden_dim) for i in range(n_channels)])
self.actvn = nn.ReLU()
self.hidden_dim = hidden_dim
if unet:
self.unet = UNet(c_dim, in_channels=c_dim, **unet_kwargs)
else:
self.unet = None
self.reso_plane = plane_resolution
self.reso_grid = grid_resolution
self.plane_type = plane_type
self.padding = padding
if scatter_type == 'max':
self.scatter = scatter_max
elif scatter_type == 'mean':
self.scatter = scatter_mean
else:
raise ValueError('incorrect scatter type')
self.pos_encoding = pos_encoding
if pos_encoding:
self.pe = positional_encoding()
def __init__(self,
c_dim=128,
dim=3,
hidden_dim=128,
scatter_type='max',
unet=False,
unet_kwargs=None,
unet3d=False,
unet3d_kwargs=None,
plane_resolution=None,
grid_resolution=None,
plane_type='xz',
padding=0.1,
n_blocks=5,
pos_encoding=False):
super().__init__()
self.c_dim = c_dim
if pos_encoding == True:
dim = 60
self.fc_pos = nn.Linear(dim, 2 * hidden_dim)
self.blocks = nn.ModuleList([
ResnetBlockFC(2 * hidden_dim, hidden_dim) for i in range(n_blocks)
])
self.fc_c = nn.Linear(hidden_dim, c_dim)
self.actvn = nn.ReLU()
self.hidden_dim = hidden_dim
if unet:
self.unet = UNet(c_dim, in_channels=c_dim, **unet_kwargs)
else:
self.unet = None
if unet3d:
#self.unet3d = UNet3D(**unet3d_kwargs)
self.unet3d = UNet3D_latent(**unet3d_kwargs)
else:
self.unet3d = None
self.reso_plane = plane_resolution
self.reso_grid = grid_resolution
self.plane_type = plane_type
self.padding = padding
if scatter_type == 'max':
self.scatter = scatter_max
elif scatter_type == 'mean':
self.scatter = scatter_mean
else:
raise ValueError('incorrect scatter type')
self.pos_encoding = pos_encoding
if pos_encoding:
self.pe = positional_encoding()
def __init__(self, dim=3, z_dim=128, c_dim=128,
hidden_size=256, leaky=False, sample_mode='bilinear', n_blocks=5, pos_encoding=False, padding=0.1):
super().__init__()
self.z_dim = z_dim
self.c_dim = c_dim
self.n_blocks = n_blocks
if pos_encoding == True:
dim = 60 # hardcoded
if z_dim != 0:
self.fc_z = nn.ModuleList([
nn.Linear(z_dim, hidden_size) for i in range(n_blocks)
])
self.conv_layers = nn.ModuleList([
nn.ConvTranspose2d(2, 6, 3, stride=2, padding=1, output_padding=1,),
nn.ConvTranspose2d(6, 12, 3, stride=2, padding=1, output_padding=1,),
nn.ConvTranspose2d(12, 24, 3, stride=2, padding=1, output_padding=1,),
nn.ConvTranspose2d(24, 48, 3, stride=2, padding=1, output_padding=1,),
nn.ConvTranspose2d(48, 96, 3, stride=2, padding=1, output_padding=1,),
])
if c_dim != 0:
self.fc_c = nn.ModuleList([
nn.Linear(c_dim, hidden_size) for i in range(n_blocks)
])
self.fc_p = nn.Linear(dim, hidden_size)
self.blocks = nn.ModuleList([
ResnetBlockFC(hidden_size) for i in range(n_blocks)
])
self.fc_out = nn.Linear(hidden_size, 1)
if not leaky:
self.actvn = F.relu
else:
self.actvn = lambda x: F.leaky_relu(x, 0.2)
self.sample_mode = sample_mode
self.padding = padding
self.pos_encoding = pos_encoding
if pos_encoding:
self.pe = positional_encoding()
def __init__(self,
dim=3,
c_dim=128,
hidden_size=256,
n_blocks=5,
leaky=False,
sample_mode='bilinear',
padding=0.1,
pos_encoding=False):
super().__init__()
self.c_dim = c_dim
self.n_blocks = n_blocks
if c_dim != 0:
self.fc_c = nn.ModuleList(
[nn.Linear(c_dim, hidden_size) for i in range(n_blocks)])
if pos_encoding == True:
dim = 60
self.fc_p = nn.Linear(dim, hidden_size)
self.blocks = nn.ModuleList(
[ResnetBlockFC(hidden_size) for i in range(n_blocks)])
self.fc_out = nn.Linear(hidden_size, 1)
if not leaky:
self.actvn = F.relu
else:
self.actvn = lambda x: F.leaky_relu(x, 0.2)
self.sample_mode = sample_mode
self.padding = padding
self.pos_encoding = pos_encoding
if pos_encoding:
self.pe = positional_encoding()