siren pytorch

SIRENA en Pytorch

Implementación de Pytorch de SIREN: representaciones neuronales implícitas con función de activación periódica

Instalar

$ pip install siren-pytorch

Uso

Una red neuronal multicapa basada en SIREN

 import torch
from torch import nn
from siren_pytorch import SirenNet

net = SirenNet (
    dim_in = 2 ,                        # input dimension, ex. 2d coor
    dim_hidden = 256 ,                  # hidden dimension
    dim_out = 3 ,                       # output dimension, ex. rgb value
    num_layers = 5 ,                    # number of layers
    final_activation = nn . Sigmoid (),   # activation of final layer (nn.Identity() for direct output)
    w0_initial = 30.                   # different signals may require different omega_0 in the first layer - this is a hyperparameter
)

coor = torch . randn ( 1 , 2 )
net ( coor ) # (1, 3) <- rgb value

Una capa de SIRENA

 import torch
from siren_pytorch import Siren

neuron = Siren (
    dim_in = 3 ,
    dim_out = 256
)

coor = torch . randn ( 1 , 3 )
neuron ( coor ) # (1, 256)

Activación sinusoidal (solo una envoltura alrededor de torch.sin )

 import torch
from siren_pytorch import Sine

act = Sine ( 1. )
coor = torch . randn ( 1 , 2 )
act ( coor )

Wrapper para entrenar en una imagen específica de altura y ancho específicos de un SirenNet determinado y luego generarla posteriormente.

 import torch
from torch import nn
from siren_pytorch import SirenNet , SirenWrapper

net = SirenNet (
    dim_in = 2 ,                        # input dimension, ex. 2d coor
    dim_hidden = 256 ,                  # hidden dimension
    dim_out = 3 ,                       # output dimension, ex. rgb value
    num_layers = 5 ,                    # number of layers
    w0_initial = 30.                   # different signals may require different omega_0 in the first layer - this is a hyperparameter
)

wrapper = SirenWrapper (
    net ,
    image_width = 256 ,
    image_height = 256
)

img = torch . randn ( 1 , 3 , 256 , 256 )
loss = wrapper ( img )
loss . backward ()

# after much training ...
# simply invoke the wrapper without passing in anything

pred_img = wrapper () # (1, 3, 256, 256) 

Modulación con código latente

Un nuevo artículo propone que la mejor manera de acondicionar una sirena con un código latente es pasar el vector latente a través de una red de avance del modulador, donde el estado oculto de cada capa se multiplica por elementos con la capa correspondiente de la sirena.

Puede usar esto simplemente configurando una palabra clave adicional latent_dim en SirenWrapper

 import torch
from torch import nn
from siren_pytorch import SirenNet , SirenWrapper

net = SirenNet (
    dim_in = 2 ,                        # input dimension, ex. 2d coor
    dim_hidden = 256 ,                  # hidden dimension
    dim_out = 3 ,                       # output dimension, ex. rgb value
    num_layers = 5 ,                    # number of layers
    w0_initial = 30.                   # different signals may require different omega_0 in the first layer - this is a hyperparameter
)

wrapper = SirenWrapper (
    net ,
    latent_dim = 512 ,
    image_width = 256 ,
    image_height = 256
)

latent = nn . Parameter ( torch . zeros ( 512 ). normal_ ( 0 , 1e-2 ))
img = torch . randn ( 1 , 3 , 256 , 256 )

loss = wrapper ( img , latent = latent )
loss . backward ()

# after much training ...
# simply invoke the wrapper without passing in anything

pred_img = wrapper ( latent = latent ) # (1, 3, 256, 256) 

Citas

 @misc { sitzmann2020implicit ,
    title   = { Implicit Neural Representations with Periodic Activation Functions } ,
    author  = { Vincent Sitzmann and Julien N. P. Martel and Alexander W. Bergman and David B. Lindell and Gordon Wetzstein } ,
    year    = { 2020 } ,
    eprint  = { 2006.09661 } ,
    archivePrefix = { arXiv } ,
    primaryClass = { cs.CV }
}

 @misc { mehta2021modulated ,
    title   = { Modulated Periodic Activations for Generalizable Local Functional Representations } , 
    author  = { Ishit Mehta and Michaël Gharbi and Connelly Barnes and Eli Shechtman and Ravi Ramamoorthi and Manmohan Chandraker } ,
    year    = { 2021 } ,
    eprint  = { 2104.03960 } ,
    archivePrefix = { arXiv } ,
    primaryClass = { cs.CV }
}