StyleGAN2Generator

class torchelie.models.StyleGAN2Generator(noise_size: int, ch_mul: float = 1, img_size: int = 128, max_ch: int = 512, mapping_lr_mul: float = 0.01)

Generator from StyleGAN2

Parameters
  • noise_size (int) – size of the input gaussian noise

  • ch_mul (float) – width adaptor, multiplies the number of channels each layer (default: 1.)

  • img_size (int) – spatial size of the image to generate (default: 128)

  • max_ch (int) – maximum number of channels (default: 512)

forward(z, mixing: bool, get_w: typing_extensions.Literal[False])torch.Tensor
forward(z, mixing: bool, get_w: typing_extensions.Literal[True])Tuple[torch.Tensor, torch.Tensor]
use_affine_input()
w_to_dict(w)
w_avg: torch.Tensor