espnet2.enh.layers.ncsnpp_utils.layers.ddpm_conv1x1

source

espnet2.enh.layers.ncsnpp_utils.layers.ddpm_conv1x1(in_planes, out_planes, stride=1, bias=True, init_scale=1.0, padding=0)

1x1 convolution with DDPM initialization.

This function creates a 1x1 convolutional layer and initializes its weights using the DDPM (Denoising Diffusion Probabilistic Models) method. The initialization is done by scaling the weights according to a specified scale parameter.

• Parameters:
  • in_planes (int) – Number of input channels.
  • out_planes (int) – Number of output channels.
  • stride (int , optional) – Stride of the convolution. Default is 1.
  • bias (bool , optional) – If True, adds a learnable bias to the output. Default is True.
  • init_scale (float , optional) – Scale for weight initialization. Default is 1.0.
  • padding (int , optional) – Zero-padding added to both sides of the input. Default is 0.
• Returns: A 1x1 convolutional layer with DDPM initialization.
• Return type: nn.Conv2d

Examples

>>> conv_layer = ddpm_conv1x1(3, 16)
>>> print(conv_layer)
Conv2d(3, 16, kernel_size=(1, 1), stride=(1, 1), bias=True)

NOTE

The weights are initialized by calling the default_init function with the specified init_scale. The bias is initialized to zero.