espnet2.enh.layers.ncsnpp_utils.upfirdn2d.upfirdn2d

source

espnet2.enh.layers.ncsnpp_utils.upfirdn2d.upfirdn2d(input, kernel, up=1, down=1, pad=(0, 0))

UpFirDn2d functions for upsampling, padding, FIR filter, and downsampling.

These functions are designed to perform a series of operations that include upsampling, applying a FIR filter, and downsampling in a two-dimensional space. The implementation is adapted from the StyleGAN2 repository (https://github.com/NVlabs/stylegan2).

• Parameters:
  • input (torch.Tensor) – The input tensor of shape (N, C, H, W) where N is the batch size, C is the number of channels, H is the height, and W is the width.
  • kernel (torch.Tensor) – The FIR filter kernel tensor of shape (KH, KW) where KH is the height and KW is the width of the kernel.
  • up (int , optional) – The upsampling factor for both dimensions. Default is 1.
  • down (int , optional) – The downsampling factor for both dimensions. Default is 1.
  • pad (tuple , optional) – A tuple of two integers specifying the padding (pad_y0, pad_y1) for height and (pad_x0, pad_x1) for width. Default is (0, 0).
• Returns: The output tensor after applying the upsampling, FIR filter, and downsampling operations. The output shape is (N, C, out_h, out_w) where out_h and out_w are calculated based on the input dimensions and the specified factors.
• Return type: torch.Tensor

Examples

>>> input_tensor = torch.randn(1, 3, 64, 64)  # Example input
>>> kernel = torch.ones((3, 3))  # Example kernel
>>> output_tensor = upfirdn2d(input_tensor, kernel, up=2, down=1, pad=(1, 1))
>>> print(output_tensor.shape)  # Output shape will vary based on input

NOTE

This function assumes that the input tensor has at least four dimensions (N, C, H, W). The kernel must be a 2D tensor.

• Raises:
  • ValueError – If the input tensor or kernel does not have the
  • expected dimensions. –